EI2N 2026 Abstracts


Area 1 - 20th International IFAC/IFIP Workshop on Enterprise Integration, Interoperability and Networking

Full Papers
Paper Nr: 7
Title:

LLM-Assisted Ontology Engineering Framework for Human-AI Work Collaboration

Authors:

Kolitha Kottagaha, Elisabeta Veliaj and Christos Emmanouilidis

Abstract: Developing application-specific ontologies for human–AI collaboration requires integrating heterogeneous reusable ontologies, domain knowledge, and cognitive-interoperability requirements. Existing Large Language Model (LLM)-assisted approaches commonly treat ontology alignment, knowledge extraction, and validation as separate tasks, with limited attention to how their outputs can be connected under human supervision. This paper presents a three-phase framework comprising ontology alignment and integration, domain specialisation, and ontology validation. In each phase, LLMs generate or classify candidate artefacts, while human reviewers retain responsibility for semantic acceptance and formalisation. The framework is instantiated in a steep-slope vineyard pilot involving human workers, AI systems, Digital Twins, and drones. Phase~1 produced a reviewed merged ontology from four reusable resources. Phase~2 produced an application-specific extension by grounding knowledge extracted from pilot documentation and stakeholder meeting transcripts against the merged ontology. Phase~3 generated competency questions from the pilot documentation and used the reviewed questions, SPARQL queries, and logical-consistency checking to evaluate the specialised ontology. All queries returned non-empty bindings against a synthetic validation knowledge graph, while HermiT detected no logical inconsistencies. The evaluation is limited to one pilot, one combination of reusable ontologies, and synthetic instances for query validation. Across all phases, human review led to the rejection or revision of generated artefacts before formalisation, indicating that LLM outputs required assessment before incorporation into the ontology. The results demonstrate how LLM assistance can connect alignment, specialisation, and validation while preserving human supervision.

Paper Nr: 9
Title:

AAS-Based Agentic AI for Automated Data Pipeline Generation in Industry 4.0

Authors:

João Miranda, António Monte Pegado, Nelson Freitas and André Dionisio Rocha

Abstract: Industrial data pipelines are essential for transforming heterogeneous shop-floor data into usable information, yet their engineering remains predominantly manual and difficult to reuse across changing manufacturing environments. Engineers must repeatedly interpret asset capabilities, redefine data transformations, configure storage and processing services, and maintain these configurations as industrial systems evolve. This lack of machine-interpretable pipeline models restricts interoperability, increases deployment effort, and limits the adaptability expected from Industry 4.0 systems. This paper presents an Asset Administration Shell (AAS) approach for representing, generating, and deploying industrial data pipelines. First, it extends existing data management AAS submodels with a reusable data-pipeline model structured around extraction, integration, preprocessing, storage, processing, and utilization stages. Second, it introduces an agentic workflow that interprets AAS-described assets through an AAS Model Context Protocol server. Third, it formulates pipeline configurations, validates execution plans with human supervision, and deploys the required software components. A containerized prototype demonstrates the creation, deployment, reuse, and extension of pipelines connecting robotic, messaging, storage, and visualization components. By treating data pipelines as first-class AAS artefacts, the proposed approach establishes a semantic foundation for more interoperable, reusable, and human-supervised industrial data engineering.

Paper Nr: 10
Title:

Supply Chain Integration, Adaptability, and AI Readiness in Nanostore Performance: A Multi-Country SEM Study of Industry 5.0 Last-Mile Logistics in Emerging Markets

Authors:

Cesar H. Ortega-Jimenez, Dany N. Sabillón Palomeque and Narciso A. Melgar-Martínez

Abstract: In the era of artificial intelligence and Industry 5.0, designing logistics sys-tems that bridge informal last-mile exchanges with lightweight analytical tools represents a frontier challenge for supply chain and logistics engineering. Nanostores—family-owned micro-retailers that play an important role in household consumer-goods distribution in Latin America and provide neigh-borhood-level retail access across parts of MENA—constitute a critical last-mile node. Their procurement and ordering processes often rely on informal channels that generate limited structured data for AI analytics. This infor-mation-processing gap can constrain visibility, forecasting, and supply-chain coordination at the last-mile level. This study examines supply chain integra-tion (SCI), adaptability, and AI readiness as antecedents of nanostore per-formance. Data were collected from 1,412 nanostores across seven countries (Honduras, Colombia, Morocco, Jordan, Saudi Arabia, Israel, and the United Arab Emirates) and analyzed using Structural Equation Modeling. Confirm-atory Factor Analysis first validated the measurement instrument before structural relationships were estimated. Results indicate positive associations between supply chain integration (β = 0.32, p < 0.001), adaptability (β = 0.26, p < 0.001), AI readiness (β = 0.30, p = 0.003), and performance. The interaction between supply chain integration and AI readiness is positive and statistically significant (β = 0.19, p = 0.022), indicating that the association between supply chain integration and performance is stronger at higher levels of perceived AI readiness. These findings are relevant to Supply Chain and Logistics Engineering, Digital Business Ecosystems in Logistics, and Supply Chain Resilience. They suggest a design hypothesis for last-mile logistics: AI tools may be more suitable when they extend existing data flows rather than require standalone applications, with attention to low data-entry require-ments, supplier-platform integration, and connectivity constraints in devel-oping-economy contexts.

Short Papers
Paper Nr: 5
Title:

An AI-Enhanced Platform for Data Integrity Risk Assessment and Action Planning in Regulated Pharmaceutical Environments

Authors:

Yessine Hammami, Mohamed Belcaid and Safa Bhar Layeb

Abstract: Data integrity is a critical prerequisite for trustworthiness, compliance, and quality in regulated pharmaceutical and digital health ecosystems. Traditional risk assessments are predominantly manual, expert-driven, and difficult to standardize at scale. This paper presents an AI-enhanced framework for data integrity risk management grounded in the Active Pharmaceutical Ingredients Committee data integrity methodology. The proposed platform integrates three components: a dynamic risk assessment module, a multi-level analytics and visualization layer, and Retrieval-Augmented Generation systems powered by large language models that deliver contextualized mitigation recommendations and on-demand expert-level regulatory guidance. Developed and tested within a large pharmaceutical organization, the framework automates data integrity governance workflows while promoting consistency in risk evaluation across organizational units. The framework was evaluated by a panel of pharmaceutical quality professionals, combining qualitative expert assessment across multiple industrial sites with structured quantitative metric evaluation. Results demonstrate strong regulatory completeness and faithfulness of generated action plans with respect to applicable regulatory requirements, alongside unanimous positive expert judgment on pertinence, applicability, and time gain.

Paper Nr: 8
Title:

Modeling of Software and Service Products with Asset Administration Shells

Authors:

Markus Rentschler, David Dietrich, Christina Bornträger and Frank Scherenschlich

Abstract: Asset Administration Shells (AAS) have become a foundational technology for interoperable digital twins in Industry 4.0. However, while substantial progress has been made in modeling physical hardware assets, standardized approaches for describing software products and software-based services remain largely underdeveloped. This position paper analyzes these gaps across existing AAS Submodel Templates and proposes a set of conceptual modeling patterns and extensions for software and services across their lifecycle. The proposed approach targets consistent representation of software features, licensing information, and functional capabilities, supporting automated discovery and semantic matching in digital marketplaces and Manufacturing-as-a-Service (MaaS) environments. As a work in progress, the proposals have not yet been formally validated in a reference AAS environment; the paper outlines the validation steps considered necessary to bridge the gap between engineering tools, service platforms, and operational systems for software and service interoperability in smart industries.

Paper Nr: 11
Title:

Socio-Cognitive AI with an Assistant Social Robot as Human Digital Twin

Authors:

Dimitra Anastasiou, Ben Gaffinet, Joseph Birardi and Yannick Naudet

Abstract: This paper presents a socio-cognitive AI framework that combines Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), ontology-based knowledge representation, and Cognitive Architectures to create a Human Digital Twin (HDT) supported by an assistant social robot. The system is designed to monitor human actions, predict intentions, and provide adaptive guidance in industrial environments to improve safety and collaboration. A cognitive social robot acts as the interface between the human operator and the HDT, using context-aware reasoning and grounded knowledge retrieval to support decision-making and interaction. As a proof of concept, the approach is demonstrated in a laboratory Lego assembly scenario in which a user receives context-aware assistance and feedback from the robot. The work establishes the technical basis for investigating human-state modelling, intention prediction, adaptive intervention, and human–CPS cognitive interoperability, while identifying systematic empirical evaluation of these capabilities as an important direction for future work.

Paper Nr: 6
Title:

We Still Need Architecture and Models: Revisiting 20 Years of GAF and GMF for Human-AI Collaboration Interoperability

Authors:

Mengjin Qu, Shihong Li and Qing Li

Abstract: As artificial intelligence continues to advance and cyber-physical-social systems become more tightly integrated, a growing number of industrial applications are taking on the characteristics of systems of systems in which people, AI, and physical assets act together. System architecture, systems methodology, and system modeling are established instruments in systems engineering, especially in system integration. They help describe a system as a whole, explain its behavior, define its boundaries, and clarify its underlying meaning. The General Architecture Framework (GAF) and the General Modelling Framework (GMF), originally developed for complex systems of systems, have evolved for nearly two decades and have acquired renewed relevance in the context of industrial intelligence. Taking the twentieth EI2N event and the twentieth anniversary of GAF as its point of departure, this study reviews the development from the stair-like architecture to the explicit separation of GAF and GMF. It then examines their relationships with international standards and representative architectures for complex systems, and discusses how model-driven approaches may connect activities across lifecycle stages. The paper concludes by outlining possible directions for applying and extending the frameworks in complex industrial-intelligence systems. The analysis suggests that GAF and GMF are moving beyond their original role as enterprise-modeling tools toward a more general architecture-and-modeling framework with interoperability potential. In this capacity, they may offer a useful basis for organizing heterogeneous elements and linking models in complex industrial-intelligence settings.