Source-linked AI summary
Digital Twins: State of the Art Theory and Practice, Challenges, and Open Research Questions
Angira Sharma, Edward Kosasih, Jie Zhang, Alexandra Brintrup, Anisoara Calinescu
TL;DR
Digital twins promise real-time monitoring, simulation, forecasting, and improved decisions, but their theory and implementation remain fragmented. This paper reviews the field, examines domain dependence and technological dependencies, and proposes a reference framework whose components support implementation and evaluation.
Problem
Digital twins lack a widely adopted universal model, making their theoretical specification, implementation, and evaluation difficult to standardize.
Method
The paper reviews theoretical and practical digital-twin work across domains, examining technological dependence, machine learning, big data, implementation gaps, challenges, and limitations.
Results
The paper defines a digital-twin reference model with essential components, including a bijective physical-asset relation, testing, security, integrated data, simulation, monitoring, and analytics.
Takeaways & Limitations
A comprehensive reference framework can clarify the essence, uniqueness, performance, and integrity of digital-twin systems across domain-dependent implementations.
Takeaways & Limitations
Digital-twin implementation can be costly and time-consuming because it requires interoperable components, real-time tools, big-data resources, and continuing investment in dependent technologies.
Abstract
from arXiv · showhide
Digital Twin was introduced over a decade ago, as an innovative all-encompassing tool, with perceived benefits including real-time monitoring, simulation and forecasting. However, the theoretical framework and practical implementations of digital twins (DT) are still far from this vision. Although successful implementations exist, sufficient implementation details are not publicly available, therefore it is difficult to assess their effectiveness, draw comparisons and jointly advance the DT methodology. This work explores the various DT features and current approaches, the shortcomings and reasons behind the delay in the implementation and adoption of digital twin. Advancements in machine learning, internet of things and big data have contributed hugely to the improvements in DT with regards to its real-time monitoring and forecasting properties. Despite this progress and individual company-based efforts, certain research gaps exist in the field, which have caused delay in the widespread adoption of this concept. We reviewed relevant works and identified that the major reasons for this delay are the lack of a universal reference framework, domain dependence, security concerns of shared data, reliance of digital twin on other technologies, and lack of quantitative metrics. We define the necessary components of a digital twin required for a universal reference framework, which also validate its uniqueness as a concept compared to similar concepts like simulation, autonomous systems, etc. This work further assesses the digital twin applications in different domains and the current state of machine learning and big data in it. It thus answers and identifies novel research questions, both of which will help to better understand and advance the theory and practice of digital twins.
1. Introduction
Digital twins are presented as virtual models intended to support real-time monitoring, simulation, forecasting, and decision-making for physical assets. Their adoption remains hindered by technological dependence, domain knowledge requirements, evolving definitions, absent standards, and implementation uncertainty.
- Motivation: Digital twins aim to provide virtual representations that support monitoring, simulation, forecasting, and timely decisions about physical assets.The motivating vision includes testing prototypes virtually, collecting real-time information, analyzing it, and anticipating problems.
- Implementation gaps: Real-time monitoring depends on IoT devices and enterprise information systems, while analytics depends on big data and machine learning tools.The paper treats these dependencies as central to DT implementation.
- Implementation gaps: DT implementation requires extensive domain knowledge because technologies must be combined for one or more physical assets.The paper compares this requirement with the domain knowledge needed to create a physical prototype.
- Implementation gaps: No universally accepted DT definition or implementation standards exist, and the DT vision continues to evolve with technology, industry, and customer needs.The absence of standards further impedes widespread design, implementation, and adoption.
- Paper objectives: The paper proposes studying these gaps through a comprehensive theoretical specification, a DT reference framework, and an implementation and evaluation methodology.Its stated contributions include examining technology dependence, domain effects, the ideal-to-practical gap, limitations, prior efforts, and DT definition and implementation.
2. Previous Work
Prior DT reviews commonly emphasize definitions, categorization, and theoretical models, while practical implementations remain underexplored and insights are often domain-specific. This review instead examines implementation challenges, practical applications, and novel questions aimed at advancing DT theory and practice.
- Existing reviews: Existing reviews often focus on specific domains or general classifications, producing insights that are difficult to transfer across domains.Examples include manufacturing, aerospace, production science, and a general comparison of 87 applications.
- Existing reviews: Practical implementations receive little attention in previous work because domain dependence and multiple technology dependencies make universal, sufficiently detailed models difficult to formulate.Prior reviews therefore emphasize paper categorization and theoretical model development more than implementation.
- Open issues: Earlier research identified needs for a broad DT definition, concrete case studies, domain-specific reference models, IoT and information-systems connections, lifecycle analysis, and data-related investigation.These questions cover both conceptual boundaries and implementation requirements.
- Novel contribution: This review adds questions about a universal reference framework, the concept-to-implementation gap, real-time machine learning and data, adoption barriers, domain effects, and implementation success.The review explicitly frames these as novel aspects and questions.
- Novel contribution: Unlike categorization-focused reviews, this work analyzes practical implementations and the root causes of challenges affecting DT theory and practice.It presents a different viewpoint on existing literature and DT implementations.
3. Digital Twin
The paper frames digital twins as synchronized digital representations of physical assets whose distinctive value comes from integrating real-time data, analytics, machine learning, and feedback. It consolidates prior definitions into elementary and imperative components, then proposes a reference framework spanning physical assets or lifecycles.
- Defining DT: Digital twin definitions vary between a final product and an entire product lifecycle, while multiple terms have delayed consensus on a unifying definition.The paper uses “asset” for the conventionally termed product and treats either a product or its lifecycle as the physical counterpart.
- Components of DT: A digital twin requires a physical asset, a digital asset, and a two-way synchronized relation carrying information between them.The physical asset may be a product or product lifecycle, and the synchronized relation is elementary rather than optional.
- Components of DT: Imperative components add IoT-based real-time collection, integrated continuous data, machine learning, data-flow security, and performance evaluation.Evaluation requires metrics and tests covering properties such as accuracy, resilience, robustness, and costs.
- Dynamic Properties: The paper identifies real-time physical connection and self-evolution as inherent properties, with self-evolution enabling real-time learning, adaptation, and feedback.The authors associate self-evolution with machine-learning approaches such as reinforcement learning.
- Dynamic Properties: The proposed reference framework allows a digital twin to represent either an asset or its production lifecycle, provided all defined components are present.The framework is presented as a synthesis of current literature and analysis.
- Different from existing technologies: Simulation, monitoring, testing, analytics, prototyping, and end-to-end visibility can be subsystems of a digital twin, but their combination with the full component set distinguishes DT.The paper states that removing any component voids the twin’s functionality and uniqueness.
3.4. How is DT di
The paper distinguishes digital twins from earlier digital and virtual concepts by emphasizing integrated, real-time synchronization and broader technology-enabled functionality. Related approaches generally exchange data manually, one way, or for only part of a product.
- Digital Model and Digital Shadow: Digital models use manual data exchange, digital shadows use one-way physical-to-digital flow, whereas digital twins fully integrate information flow and reflect the physical object’s actual state.These distinctions concern synchronization direction and whether the digital representation tracks current physical conditions.
- Semantic Virtual Factory Data Model: Semantic Virtual Factory Data Models represent factory entities as data models, while digital twins add real-time synchronization.The comparison is framed around data-model status versus synchronized operation.
- Product Avatar: Product Avatars manage distributed product information without feedback and may represent only parts of a product.This limits their correspondence with the paper’s broader digital-twin concept.
- Digital Product Memory: Digital product memories sense and capture information about a specific physical part, making them an instantiation of a digital twin.The paper presents digital product memory as an extension of semantic or digital product memory.
- Intelligent Product and Holons: Digital twins extend intelligent products through IoT, big data, and machine learning technologies that were absent from the earlier concept.Holons are described as an earlier computer-integrated manufacturing basis for these related technologies.
4. Existing Models
Existing digital-twin models span general architectures and domain-specific implementations, but practical models remain difficult to compare because implementation details, evaluation methods, and cross-domain requirements vary. The literature indicates that digital twins retain common components while their feasibility, challenges, and implementation depth depend strongly on domain.
- Theoretical models: Most digital-twin architectures are domain-specific, while only a few attempt generalization across sectors.
- Theoretical models: Digital twins are debated as representations of products or entire product lifecycles, with both interpretations supported when all required components are present.
- Practical models: Practical implementations remain scant and increasingly vague for large, complex systems, with methodology also depending on domain.
- Domain dependence: Aerospace digital twins can replicate thermal, mechanical, and acoustic conditions beyond laboratory testing, but aircraft-scale component and software interoperability make unique twin creation challenging.
- Machine learning: Machine-learning use in digital twins is described as limited, despite its relevance to self-evolution, resilience, and analytics.
- Domain dependence: Domain dependence affects implementation feasibility, data collection, realism, optimization complexity, component compatibility, and regulation or standards.
1. Investing in DT
Signify Philips is exploring digital twins for lighting, including emergency services, real-time monitoring, and predictive maintenance.
- Signify Philips is exploring digital twins for lighting with emergency services, real-time monitoring, and predictive maintenance.
2. Providing DT as a service
Several companies provide digital-twin technologies or use them in sector-specific systems, including healthcare, ports, power grids, racing, and software services. The reviewed examples do not establish global-scale implementation, and machine-learning use is often unclear.
- Healthcare: Philips provides digital-twin technology for healthcare systems to detect early technical warnings in MRI and CT equipment.
- Sector applications: IBM, Siemens, and other providers apply digital twins to port monitoring, PLM, power grids, and Formula 1 racing.
- Software services: Dassault Systèmes, AnyLogic, Ansys, Visualiz, PwC, Bosch, SAP, Azure, Oracle, and GE provide digital-twin software or related services.
3. Using DT for own use
Organizations use digital twins for real-time monitoring, simulation, optimization, and predictive functions, but published evidence rarely exposes implementation depth or evaluation metrics. Machine learning and big-data integration remain technically difficult because of synchronization, scale, dimensionality, and domain-specific requirements.
- Organizational use: DHL implemented a real-time digital-twin supply chain for a Tetra Pak warehouse, while BP used APEX to create virtual copies of production systems.
- Evaluation: Existing implementations often omit frameworks and quantitative evaluation, making their success difficult to assess despite claims involving monitoring, simulation, optimization, and data handling.
- Machine learning: Real-time machine learning distinguishes a digital twin from a simulator or monitoring tool by using incoming data to predict future asset behavior.
- Machine learning: A petrochemical proof of concept used multiple machine-learning algorithms, but its implementation methodology and real-time feedback loop were not described.
- Implementation challenges: Digital-twin integration requires high-fidelity synchronization, joint multi-objective optimization, domain knowledge, and compatible IoT, data, and machine-learning components.
- Big data: Digital-twin data challenges include high dimensionality, mismatched time frequencies, fragmentation, storage and preprocessing demands, and synchronization across components.
- Data integrity: Blockchain may improve data integrity but adds complexity to digital-twin data handling and component optimization.
5. Industries where DT can be majorly beneficial
Digital twins are especially beneficial where physical prototyping is costly, extreme testing is impractical, or real-time monitoring is needed. Despite requiring multiple technologies and expertise, they can reduce costs, shorten design cycles, and support timely mitigation.
- Digital twins benefit industries where physical prototypes are expensive, resource-intensive, or time-consuming, including aerospace, supply chain, and manufacturing.
- Digital twins can simulate extreme tests that are difficult or impossible to perform in laboratory settings, particularly in aerospace and prognostics and health management.
- Digital twins support industries requiring real-time monitoring and mitigation planning.
- A one-time investment in digital twin technology can enable cost reductions, shorter design cycles, and savings in prototyping time and resources.Cost reduction may serve as a performance metric for profit-oriented companies.
6. Current Challenges and Limitations in DT
Digital twin implementations face technical challenges in handling large, asynchronous, multi-source data and achieving high-fidelity synchronisation. They also require costly integration of simulation, optimisation, interoperability, and evolving supporting technologies.
- Challenges: Digital twins must process high-dimensional, time-series, multimodal, and multisource data collected from numerous IoT devices.Mismatched collection frequencies can fragment data and create time lags.
- Challenges: High-fidelity two-way synchronisation is difficult at large scale because it requires substantial resources and high-stream IoT connectivity.
- Challenges: Digital twin simulation software requires programmers, developers, and domain experts, while simulation-based optimisation remains difficult to integrate.
- Limitations: Digital twin limitations include cost for short-lived projects, implementation complexity, and the need to update dependent technologies over time.
7. Open Research Questions and Future Steps
The paper identifies unresolved questions about quantitatively evaluating digital twins and correcting data or model errors, then proposes formal definitions, IoT standards, and regulations as future steps.
- Open Research Questions: Digital twin performance requires quantitative, domain-dependent and domain-independent metrics, including uncertainty quantification and confidence levels for outputs.Quantitative self-assessment is also identified as necessary for self-evolving digital twins.
- Open Research Questions: Research is needed to rectify noisy or inaccurate data and model errors because incorrect inputs can produce incorrect predictions or test results.The paper notes that human oversight remains important when deciding whether to follow a digital twin recommendation.
- Future Steps: A formal digital twin definition is proposed to address the lack of consensus and support a universally accepted concept.
- Future Steps: IoT standards for data capture, sharing, real-time synchronisation, and monitoring are proposed to facilitate digital twin acceptance and adoption.
- Future Steps: Regulations and security mechanisms are proposed to address data-sharing concerns across collaborators and industry partners.Federated learning is identified as a promising related direction.
8. Conclusions
The paper defines a reference model for digital twins from prior work and identifies components that capture their essence and support system evaluation. It concludes that digital twins combine multiple tailored subsystems but remain constrained by technical and domain-dependent challenges.
- The proposed digital twin reference model addresses the absence of a widely adopted universal model.
- The model includes a bijective relation between a digital twin and its physical asset, along with testing and security components for evaluating performance and integrity.
- Digital twins combine simulation, autonomy, agent-based modeling, machine learning, prototyping, optimisation, and big data into a configurable tool.Subsystems can be tailored and prioritised according to application-domain needs.
- Digital twin advancement remains subject to technical and domain-dependent challenges because the technology depends on IoT, machine learning, and data.The paper identifies seamless integration of these counterparts as producing a powerful and efficient digital twin.