Cognitive and Semantic Foundations for Resilient Digital Twins in Smart Industries

Abstract

Digital Twins (DTs) have become a cornerstone of Industry 4.0 and Industry 5.0, enabling the representation, monitoring, and control of complex industrial assets and processes. However, current DT implementations remain largely data-driven and often lack the capacity to formalize knowledge, reason under uncertainty, and dynamically adapt to disruptions. This Open Invited Track (OIT) proposes to advance the paradigm of Cognitive Digital Twins (CDTs) by integrating knowledge formalisation, cognition, semantics, and artificial intelligence approaches, with a particular focus on semantic interoperability as a foundation for scalable and trustworthy collaboration across heterogeneous systems. By integrating neural network architectures including network-based language models, and hybrid AI methods into DT simulation environments, CDTs can achieve human-like reasoning, improved resilience to accidents and perturbations but also to cyber threats,  and enhanced decision support in smart industries. This track will provide a platform for interdisciplinary contributions that combine control theory, semantic technologies, AI, cyber security and industrial engineering to shape the next generation of intelligent, human-centric, and resilient industrial ecosystems.

Detailed Description of the Topic

1. Background and Motivation

The vision of Industry 4.0 and the transition toward Industry 5.0 emphasize intelligent, adaptive, and human-centric manufacturing ecosystems. DTs have emerged as enabling technologies, allowing real-time representation and simulation of physical systems within cyber-physical environments. Despite their adoption across manufacturing, logistics, and critical infrastructures, current DT approaches are limited by:

• reliance on data-centric models without structured semantics;

• weak knowledge formalisation for reasoning and interoperability;

• insufficient resilience to disruptive events, uncertainty, and system failures.

International frameworks such as the ISO 23247 Digital Twin Framework, IEEE 2801, and the NIST Cyber-Physical Systems (CPS) Framework stress the need for semantic consistency, trustworthiness, and resilience. Addressing these challenges requires embedding cognition and semantic interoperability at the core of DT architectures, enabling the emergence of Cognitive Digital Twins (CDTs) capable of understanding, reasoning, and adapting.

2. Research Gaps

A first critical research gap lies in the absence of standardized semantic models that can support Digital Twins (DTs) in achieving interoperability across domains and platforms. Although several industrial and academic initiatives have proposed ontologies and knowledge representation frameworks, they remain fragmented, domain-specific, or lacking alignment with international standards. This limits the capacity of DTs to exchange, interpret, and reuse knowledge seamlessly across heterogeneous cyber–physical systems, supply chains, and smart industry applications. The absence of robust semantic interoperability mechanisms hinders scalability and prevents the realization of truly interconnected industrial ecosystems.

A second gap concerns the limited integration of cognitive models into DT architectures. Current DTs primarily rely on data-driven or physics-based models, which, while effective for monitoring and control, lack the ability to emulate human-like reasoning, situation awareness, and intent recognition. Without such cognitive capabilities, DTs cannot dynamically interpret complex contexts, anticipate operator needs, or adapt their behavior in a way that mirrors human decision-making processes. Embedding cognition is essential to move beyond static data replication toward intelligent, adaptive, and context-sensitive DTs capable of collaborative reasoning with humans and autonomous systems.

A third research challenge lies in the limited and fragmented integration of neural network-based language models into Digital Twins (DTs). While large language models (LLMs) represent the most advanced specialization of this family, capable of contextual understanding and semantic enrichment, other forms such as small language models or functional domain-specific models also offer valuable opportunities for lightweight, targeted, and efficient applications. Despite these advances, the systematic incorporation of neural network architectures for language processing into DT frameworks remains underexplored, leaving open questions regarding their reliability, explainability, scalability, and industrial applicability. While the Language Models have demonstrated unprecedented abilities in contextual understanding, semantic enrichment, and natural language interaction, their systematic integration into industrial DT frameworks remains nascent. Similarly, neural networks, particularly in deep and hybrid forms, offer powerful tools for predictive modeling, anomaly detection, and adaptive control. However, combining these techniques with formal DT structures raises questions of reliability, explainability, and computational feasibility that have yet to be fully addressed. Bridging this gap requires innovative approaches that blend symbolic reasoning, statistical learning, and semantic grounding.

Finally, there is an insufficient development of methodologies to ensure resilience in DT-enabled systems, especially in the context of complex supply chains, manufacturing networks, and critical infrastructures. Current resilience approaches often focus narrowly on redundancy or recovery mechanisms, without integrating predictive and cognitive features that would allow DTs to anticipate disruptions, mitigate cascading failures, and support proactive decision-making under uncertainty. Developing resilient Cognitive Digital Twins therefore requires advancing simulation methods, risk-impact modeling, and adaptive control strategies that can maintain system functionality in the face of accidents, cyberattacks, or large-scale disruptions.

3. Cognitive and Semantic Enhancements to Digital Twins

The proposed Open Invited Track focuses on advancing Digital Twins (DTs) along four complementary axes that together define the foundations of Cognitive Digital Twins (CDTs).

The first axis concerns knowledge formalisation and semantic interoperability, which represent the backbone of intelligent DTs. Ontologies, knowledge graphs, and formal logic-based models provide the means to structure and integrate the wealth of data that DTs must handle. By embedding semantic layers, DTs are able not only to store and process data, but also to understand the context and meaning behind it. Achieving semantic interoperability across heterogeneous cyber–physical systems and industrial ecosystems is a major challenge, as DTs must often operate across organizational, technological, and disciplinary boundaries. To address this challenge, research must establish robust semantic models and alignment mechanisms that guarantee consistency of meaning while ensuring compliance with international standards such as ISO 23247 for digital twin frameworks, IEEE interoperability guidelines, and the NIST CPS Framework. Such standardization efforts will make it possible to move from isolated DT applications toward large-scale, networked ecosystems of interoperable Cognitive Digital Twins.

The second axis is dedicated to cognition and artificial intelligence, which provide the intelligence layer enabling DTs to reason, adapt, and interact. Embedding cognitive architectures inspired by human perception, learning, and decision-making allows DTs to move beyond descriptive and predictive models toward systems that are capable of contextual interpretation and proactive behavior. Large language models (LLMs) can enrich DTs with natural language interaction capabilities, enabling operators to query and communicate with DTs in human language, while also extracting knowledge from unstructured sources such as technical documentation or maintenance logs. Neural network architectures, in parallel, provide robust tools for predictive modeling, anomaly detection, and adaptive simulation, allowing DTs to capture dynamic and nonlinear system behaviors. The combination of symbolic reasoning with sub-symbolic methods, often referred to as hybrid AI, offers a pathway toward explainable and trustworthy CDTs that can both perform complex computations and justify their outputs in ways that are understandable to humans. Ensuring scalability is critical, as CDTs must be capable of handling not only the complexity of individual assets but also the integration of thousands of interconnected DTs operating across extended industrial ecosystems.

The third axis emphasizes simulation and decision support, which remain central to the role of DTs in industrial contexts. Traditional simulation methods are being expanded with advanced computational approaches that integrate risk management, operational optimization, and predictive analytics. Agent-based and hybrid simulation models are particularly promising for representing multi-scale industrial processes where local interactions can have global consequences. Incorporating machine learning into simulation environments enhances the ability to model uncertainty, adapt to evolving conditions, and generate actionable insights in real time. Resilient simulation methods further extend these capabilities by incorporating accident management, recovery strategies, and adaptive control, ensuring that DTs do not simply replicate physical systems but actively support decision-making in complex, uncertain, and disruptive scenarios.

The fourth axis addresses resilience in smart industries, which is a defining requirement for the next generation of DTs. Resilient CDTs must be capable of anticipating potential disruptions, absorbing their immediate impact, and recovering functionality rapidly to minimize systemic damage. This includes the design of DTs that are not only technically robust but also aligned with the socio-technical principles of Industry 4.0 and Industry 5.0, where human operators, intelligent machines, and complex networks collaborate. Human-in-the-loop collaboration becomes particularly relevant, as resilience strategies require joint human–machine decision-making, trust, and transparency. Designing trustworthy DT systems implies aligning autonomous decision-making with principles of explainability, accountability, and human oversight, ensuring that operators remain confident in relying on CDT outputs. By integrating resilience as a core design principle, CDTs will be positioned to support not only technical efficiency but also long-term sustainability and safety in smart industries. Resilience approaches must therefore be scalable, ensuring that methods effective at the level of a single machine or process can be extended to entire factories, global supply networks, or critical infrastructures.

Expected Contributions

The track welcomes contributions in the form of regular, survey, discussion, and dissemination papers on (but not limited to) the following themes:

• Semantic Interoperability & Knowledge Formalisation

  • Ontology-based and knowledge graph-driven DTs for multi-domain integration.

  • Semantic alignment methods for CPS, IoT, and industrial platforms.

  • Standards and frameworks for interoperability (ISO, IEEE, NIST, EU initiatives).

  • Federated knowledge models enabling interoperability across distributed DT ecosystems.

  • Methods for semantic data fusion and context-aware data management.

  • Scalable semantic frameworks enabling interoperability across distributed DT networks.

• Cognition & Artificial Intelligence in DTs

  • Cognitive architectures for DT reasoning, learning, and decision-making.

  • LLM-based methods for contextualization, natural language interfaces, and semantic enrichment.

  • Neural network architectures for real-time prediction, anomaly detection, and adaptive control.

  • Hybrid symbolic–connectionist AI for trustworthy CDTs.

  • Explainable AI approaches to increase transparency of CDT decision-making.

  • Multi-agent cognitive systems for cooperative DT behaviour.

• Simulation & Optimization

  • Agent-based, discrete-event, and hybrid simulations for resilience analysis.

  • Integration of machine learning and optimization in DT simulation loops.

  • Simulation of complex adaptive systems (e.g., manufacturing, logistics, healthcare).

  • Real-time simulation and digital-physical co-simulation for industrial control.

  • Multi-scale simulation frameworks linking component-level to system-level behaviors.

  • Simulation-assisted planning and optimization of resource allocation in smart industries.

  • Scalable simulation methods for large-scale, real-time industrial ecosystems.

• Resilience & Human-Centric Design

  • Resilient CDT approaches for accident prediction, risk assessment, and recovery.

  • Human-in-the-loop simulation for collaborative decision-making.

  • Trust, transparency, and explainability in CDT-enabled operations.

  • Adaptive control strategies for resilience under cyber–physical disruptions.

  • Socio-technical resilience frameworks combining technical and organizational dimensions.

  • Approaches to foster human trust, acceptance, and usability of CDTs.

  • Frameworks for scalable resilience strategies applicable across local and global industrial contexts.

• Applications and Case Studies

  • Smart manufacturing, production systems, and supply chain networks.

  • Critical infrastructure (energy, transport, water, healthcare, communications networks and services including 5G).

  • Sustainability-oriented DTs for circular and green industries.

  • Cross-domain applications linking Industry 4.0/5.0 with societal resilience.

  • Aerospace and defense applications with high requirements for safety and resilience.

  • Smart cities and urban infrastructures integrating DTs for mobility, energy, and public services.

  • Healthcare DTs for personalized medicine, hospital logistics, and resilient care delivery.

  • Agriculture and food systems using CDTs for sustainable production and supply resilience.

Contact

Dr. Mario LEZOCHE (mario.lezoche@univ-lorraine.fr)