IN4PL 2026 Abstracts


Area 1 - Industry of the Future

Full Papers
Paper Nr: 21
Title:

Optimization through Digital Twins: Algorithms under Model Uncertainty

Authors:

Christine Markarian, Mirco Antona, Sarah Di Grande and Salvatore Cavalieri

Abstract: Modern decision systems increasingly rely on digital twins and simulation models to guide optimization and control. In such settings algorithms do not interact directly with the real environment but compute decisions using a model that approximates reality. While digital twins are widely adopted in practice, the algorithmic consequences of modelmediated decision-making remain largely unexplored from a theoretical perspective. This paper introduces a framework for analyzing algorithms that operate under model uncertainty. Decisions are computed using a model environment, while performance is evaluated in the real environment. To quantify the impact of model mismatch, the Model Uncertainty Ratio is defined as a worst-case measure comparing the performance of a model-mediated algorithm with the optimal decision obtainable when the real environment is known. The framework is instantiated on a scheduling problem. Upper and lower bounds are derived that relate distortions between the model and the real environment to the resulting degradation in decision quality. The analysis shows that even small discrepancies between the two environments may lead to multiplicative performance loss, revealing structural sensitivities that are not captured by classical optimization analyses. Overall, the framework provides an algorithmic perspective on optimization through digital representations of complex systems and suggests new directions for studying algorithms that operate under model uncertainty.

Paper Nr: 26
Title:

Buffering and Fatigue-Aware Sequencing in Multi-Model Assembly Lines: A Simulation Study

Authors:

Matteo Cais, Veronica Schneider and Antonella Meneghetti

Abstract: Industry 5.0 promotes a human-centric perspective in manufacturing system engineering, encouraging production strategies that balance operational efficiency with worker well-being. In assembly lines, operators are still extensively involved in manual activities, making human factors a source of variability in system performance. Worker fatigue and experience significantly influence task execution times and affect overall productivity. This paper presents a hybrid discrete-event and agent-based simulation model of a multi-model assembly line, in which task processing times dynamically evolve according to operator fatigue accumulation and experience levels. The proposed model enables the analysis of production systems under more realistic operating conditions by accounting for the interaction between human factors and system performance. The study focuses on two main strategies: buffer adoption and the implementation of different product sequencing policies, including fatigue-aware sequencing. Simulation experiments involve different production scenarios, in order to evaluate their impact on throughput, blocking, starvation, and fatigue evolution during the shift. The results show that buffer availability is the key factor in the trade-off between throughput and ergonomic well-being. Buffers increase production capacity but eliminate involuntary recovery breaks, raising fatigue levels. Fatigue-aware sequencing outperforms static policies by simultaneously improving throughput and reducing average fatigue, but requires decoupling points. The findings highlight the importance of integrating the human-centric perspective into assembly line analysis and decision-making.

Paper Nr: 28
Title:

Explicit Linking Structures for the Processing of Operational Time Series Data in AAS-Based Digital Product Passport Environments

Authors:

Israt Nowshin, Dirk Schöttke, Stephan Schäfer, Daniel Büttner, Alexander Gordt and Felix Strohmeier

Abstract: Digital Product Passports (DPPs) implemented in Asset Administration Shell (AAS) environments can represent product information, technical reference data, semantic identifiers, and operational time series data in a common digital context. However, availability alone does not enable reproducible processing. Operational measurements must first be explicitly linked to asset context, asset identification, semantic identification, measurement object, technical reference parameter, transformation reference, validity context, and version context. This paper proposes an explicitly resolvable linking structure for AAS-based Digital Product Passport environments. It combines identity-based asset mapping, ECLASS-supported semantic identification, reference-parameter resolution, transformation-reference resolution, syntactic and semantic validation, and change-aware re-resolution. The structure is illustrated in an exemplary motor-based Operational Technology (OT) implementation context with CANopen-connected field devices in which an OT-side asset is mapped to its AAS context and linked with measurement, reference, transformation, validity, and version information to derive a valid relation instance for reference-based normalization. The contribution lies in defining the preconditions under which reference data and operational time series data become semantically identifiable, context-bound, reference-based, and further processable. Reference-based normalization is treated as a preparatory transformation step, not as a health index, fault diagnosis method, or predictive maintenance model.

Paper Nr: 31
Title:

Structural Conditions of Operational Usability of DPP-Relevant Information Objects in AAS-Based Environments

Authors:

Daniel Büttner, Dirk Schöttke, Israt Nowshin, Stephan Schäfer, Felix Strohmeier and Alexander Gordt

Abstract: Digital Product Passport (DPP) initiatives rely on structured, machine-readable product and asset information, for which the Asset Administration Shell (AAS) provides a relevant representation basis. However, the presence of DPP-relevant information in AAS-based environments does not by itself ensure that information objects remain operationally usable in later role, purpose, and life-cycle contexts. In this paper, operational usability refers to the property of an information object to be functionally usable within a given role, purpose, and life-cycle context. Its preservation across context shifts is termed cross-context continuity of use. Operational usability is understood in an information-structuring sense and is distinct from HCI usability as defined in ISO 9241 and from general data quality. The paper develops an analytical lens for examining the conditions of information structuring under which DPP-relevant information objects remain available, interpretable, subject to controlled updatability, and reusable in subsequent role, purpose, or life-cycle contexts. The lens is organised around four analytical axes: availability, interpretability, controlled updatability, and reusability. These axes are mirrored by provision, interpretation, update, and use discontinuities, while auditability is treated as a cross-cutting condition concerning source, originating context, responsible instance, change event, temporal reference, and validity state. A working classification of DPP-relevant information objects and an illustrative application context from control cabinet manufacturing demonstrate how the lens can support structured analysis of cross-context information use. The contribution clarifies information-structuring conditions that preserve cross-context continuity of use in AAS-based DPP environments and provides a vocabulary for diagnosing losses of operational usability in later use contexts.

Paper Nr: 35
Title:

Bridging LCA ISO Standards and Industrial Practice: An IDEF0 Process Model for Life Cycle Assessment

Authors:

Michael Adjei Mensah, Giuseppe Berio, Mounira Harzallah and Yasamin Eslami

Abstract: To achieve environmental performance optimization in industrial production and logistics, Life Cycle Assessment (LCA) as a key quantification method must be internalized as a fundamental design parameter. However, the integration of LCA into enterprise models is currently hindered by the nature of ISO 14040 and ISO 14044 standards as the primary reference for the LCA framework. The main problem of such standard is that it has a very high degree of complexity and could appear fragmented, making it a challenge for practitioners to follow, coupled with isolated Input-Process-Output (IPO) diagrams that withhold an overall vision. In addition, the standard leaves implicit operational activities, governing constraints and resources. This paper models the LCA procedure prescribed by these standards as an IDEF0 function model and uses its Input, Control, Output and Mechanism (ICOM) semantics to ask whether an explicit model can refine the standard's general activities and restructure their ordering to bring them closer to practice. As refinement,it decomposes each phase in operational sub-activities the standards leave implicit and makes explicit the controls and mechanisms for better reproducibility and traceability, and the selection of software and database, a decisive step no standard explicitly mentions. Reviewed with industrial partners and instantiated with one case study, the model shows its usage for bridging environmental assessment and industrial production system.

Paper Nr: 36
Title:

Procedural Correctness in AI-Supported Maintenance Troubleshooting: A Comparison of Embedding-Based and Hybrid Retrieval

Authors:

Wessel J. Reuvers, Guillermo Gil de Avalle, Laura Maruster and Christos Emmanouilidis

Abstract: Maintenance troubleshooting requires shop-floor operators to locate and follow the correct step within a structured diagnostic procedure, where an action that is correct in one branch is wrong in another. Assistants built on large language models use retrieve-and-generate frameworks to extract relevant steps from supporting documentation to generate operator advice. However, retrieval based on semantic similarity alone can return content that is topically relevant, yet drawn from the wrong document or procedure, thus affecting procedural correctness and answer quality. This study compares standard similarity-based retrieval against an approach that augments it with the graph structure of the domain documentation and procedures, referred to as ‘hybrid’ retrieval. Maintenance standard operating procedures from an operational production facility were used to construct queries at different levels of clarity. Retrieval was assessed at document, branch, and node level, and the resulting advice was assessed for grounding and action correctness. The hybrid approach improved exact-step retrieval, ranking quality, and the precision of the retrieved context, while similarity-based retrieval retained broader document and branch coverage. These results offer insights into the complementarity of these retrieval strategies in industrial maintenance applications.

Paper Nr: 42
Title:

A Spatial Augmented Reality Cyber-Physical System for Workstation Operations: Guiding Operators toward Value-Added Activities

Authors:

Jean-Thomas Sexton, Michael Morin, Éloïse Prévôt and Jonathan Gaudreault

Abstract: Fully automated, lights-out production remains uncommon even in highly automated industries; human operators often remain essential for manual assembly. Yet, instructions from manufacturing execution systems (MESs) are still commonly delivered through graphical user interfaces (GUIs), forcing operators to mentally map abstract instructions onto the physical workspace. Because much of this cognitive and physical effort is non-value added (NVA), better MES workstationlevel interfaces are needed. To address this issue, we introduce a spatially augmented manufacturing execution system (SA-MES), which upgrades the conventional MES execution interface into a sensor-driven and projection-based cyber-physical system. SA-MES uses workspace sensing and projection-based spatial augmented reality to deliver MES guidance directly where actions are performed. The system visually highlights the parts to be manipulated, guides execution in real time, assesses task completion, triggers shop-floor support when parts are missing, and dynamically reorders pending tasks during waiting times. We posit that this transformation reduces NVA activities during manual assembly. To test this claim, we conducted a controlled sequential assembly experiment with 32 participants, comparing SA-MES with a conventional GUI-based MES. In a first scenario without perturbation, SA-MES reduced median completion time from 227.23 s to 59.39 s. For a second scenario where an assembly part was missing, SA-MES reduced median total task time from 515.62 s to 58.26 s. Moreover, 8 out of 15 GUI participants made at least one placement error, but all SA-MES participants completed the task correctly. These results show that SA-MES reduces execution overhead and improves robustness to local assembly disturbances, suggesting that MES execution interfaces can move beyond passive instruction delivery toward active cyber-physical support for human assembly work.

Paper Nr: 43
Title:

Towards Cognitive Digital Twins: A Neuro-Symbolic Architecture for Real-Time Reasoning in Human-Centric Cyber-Physical Production Systems

Authors:

Marcelo F. F. Calixto, Anderson L. Szejka, Hervé Panetto and Eduardo F. R. Loures

Abstract: Digital Twins in cyber-physical production systems provide synchronised representations of physical assets for monitoring and predictive analytics, but most implementations remain data-driven and cannot interpret an operational state against engineering constraints, safety regulations or maintenance procedures. This paper proposes a neuro-symbolic Cognitive Digital Twin (CDT) architecture that embeds a symbolic reasoning layer within the Digital Twin inference loop. A transformer-based model maps joint-level sensor streams from a Universal Robots UR5 manipulator into a shared 64-dimensional latent space; a knowledge graph reasoner, grounded in a robot health management ontology and ISO 10218-1 safety constraints, projects ontology entities into that same space and emits validated decision records carrying an explicit clause reference, the affected joint indices and a prescribed corrective action with an urgency class. The architecture is positioned within the Human-Centric Cyber-Physical Production System (HCCPPS) framework and validated on real UR5 controller data from the NIST Degradation Measurement dataset, augmented by a physically grounded fault injection framework producing 8,910 labelled windows across five health classes. The evaluation reports 95.79% classification accuracy, 41 ISO 10218-1 Clause 5.8 violations identified by the symbolic layer on windows an XGBoost baseline classified as normal, and complete six-field explanation records for all 1,782 test decisions; a rule-based simulation involving no human participants verifies that the human-in-the-loop path routes and logs decisions correctly. The contribution is stated in capability terms — the symbolic layer supplies normative justification that a data-driven twin structurally cannot produce. Validated operator trust and naturally occurring fault data are explicitly outside the scope of this evaluation.

Paper Nr: 48
Title:

AAS Modelling and LLM-Enhanced Explainable AI for Zero-Defect Manufacturing for Multi-Stage

Authors:

Nelson Freitas, Raul Mestre Dinis, Tiago Gouveia, Antonio Monte Pegado, Miguel Arvana, Andre Dionisio Rocha, Jose Barata and Bruno Silva

Abstract: The integration of Artificial Intelligence (AI) into modern manufacturing is rapid-ly accelerating, driving advancements in production quality, predictive mainte-nance, and the dynamic allocation of resources. However, the rationale behind AI-driven decisions is often hidden, particularly for non-AI experts such as shop-floor operators. This "black-box" problem is severely exacerbated in com-plex, multi-stage production processes where data and automated decisions cas-cade across interconnected systems. To address this challenge, this paper presents a framework, demonstrated through a practical laboratory use case from the ZDM4MS project, that utilizes eXplainable AI (xAI), Large Language Models (LLMs), and the Asset Administration Shell (AAS) novel description. The re-sults detail the creation of a standardised AAS template specifically designed to accommodate xAI models. Furthermore, an interactive interface is introduced, translating complex technical and graphical outputs into accessible natural lan-guage explanations. By enabling dynamic querying, the system empowers users to ask specific questions and request tailored explanations, ultimately fostering transparency, trust, and human-in-the-loop decision-making within multi-stage manufacturing environments.

Paper Nr: 49
Title:

Towards Smart Food Factory: A RAMI 4.0 and Asset Administration Shell Based Digitalization Approach for Food Processing

Authors:

Miguel Arvana, Catia Carraca, Nelson Freitas, Andre Dionisio Rocha and Jose Barata

Abstract: The agri-food processing sector faces significant challenges related to process variability, strict quality and food safety requirements, heterogeneous legacy equipment, and the limited integration of digital technologies. Although the Reference Architectural Model Industrie 4.0 (RAMI 4.0) and the Asset Administration Shell (AAS) have been widely adopted in manufacturing, their application to food processing remains limited. This work proposes a RAMI 4.0-compliant digitalization approach for food processing systems based on three fundamental dimensions: resources, stages, and products. Based on these dimensions, standardized AAS templates are defined to create consistent and interoperable digital representations, supported by a layered architecture that operationalizes RAMI 4.0 through continuous data acquisition, information management, and standardized asset integration. This approach supports the progressive digitalization of legacy food processing systems through a retrofit strategy and was validated during the drying stage of an industrial charcuterie production process. The experimental validation enabled continuous monitoring of temperature, humidity, and weight, allowing process characterization, the identification of recurring patterns, and the definition of reference functions for future analysis. The results suggest that legacy food processing systems can evolve into interoperable digital environments through continuous process monitoring, standardized asset representation, data integration, and synchronized Type 2 AASs. These capabilities improve process visibility and traceability, facilitate the transition from empirical operation to data-driven decision-making, and lay the digital foundation for future intelligent services supporting the realization of Smart Food Factory environments.

Paper Nr: 50
Title:

A Hub-and-Spoke Drone Network for Time-Critical Medical Logistics in Dense Urban Areas

Authors:

John Alejandro Rondón Gamboa and Jairo R. Montoya-Torres

Abstract: Time-critical hospital logistics in dense urban areas is affected by road congestion, uncertain travel times and the need to preserve the integrity of biomedical components. This paper proposes and evaluates the Medical Air Support Chain (CSAM), an intelligent drone-based logistics network designed for inter-hospital transportation in the Greater Montreal Area. The study contributes to intelligent logistics and Industry 4.0 by integrating autonomous aerial mobility, simulation-based assessment and risk-aware network design for critical healthcare supply chains. The proposed model combines a hub-and-spoke architecture, urban air mobility principles, UTM-based traffic coordination, risk-aware operational planning and safety management requirements for BVLOS drone operations. The proposed network connects three strategic healthcare facilities in the city of Montréal, and the evaluation compares the proposed drone-enabled system with conventional ground transportation using operational, economic and safety-related indicators. Results reported in the case study show an estimated 41% improvement in logistics efficiency. The economic assessment suggests that, despite higher initial investment, lower operating costs and the value of medical time saved may support financial viability over a 36-month horizon.

Paper Nr: 52
Title:

A Two-Stage Decision Optimization Framework for Production Order Batching

Authors:

Hou Zeqiang, Ge Bingfeng, Zhu Jianghan and Li Jiaojiao

Abstract: This study proposes a two-stage decision optimization framework for the production order batching and layout optimization problem of square sheets using the benchmark dataset from Problem B of the 2022 Huawei Cup China Graduate Mathematical Contest in Modeling. The framework decomposes the problem into two sequential stages. In the first stage, three heuristic batching strategies, namely order-based, sequential, and similarity-based batching, are developed to partition production orders under practical production constraints. The similarity-based strategy groups orders according to material similarity, where orders with more than 30% material overlap are merged whenever the constraints are satisfied. In the second stage, a mixed-integer linear programming model is applied to optimize the layout of each batch with the objective of minimizing the number of sheets used. Computational experiments solved by CPLEX demonstrate that the order-based and sequential batching strategies achieve similar sheet utilization of approximately 55%, whereas the similarity-based batching strategy consistently improves sheet utilization to over 67%. Overall, the proposed framework effectively integrates production order batching and layout optimization, and the similarity-based batching strategy achieves superior solution quality and material utilization.

Paper Nr: 57
Title:

Data-Driven Seasonal Performance Evaluation of Heat Pump Integration for Cold Thermal in Industrial Whey Processing

Authors:

Beatrice Marchi, Natalya Lysova, Federico Solari, Simone Zanoni, Alessandra Neri, Damiana Chinese, Maria Grazia Marchesano, Enrico Cagno and Guido Guizzi

Abstract: Industrial dairy processing relies heavily on electrically-driven chillers to meet the cold thermal energy demand of whey cooling and concentration, a major contributor to the sector’s electricity consumption. One year of quarter-hourly consumption data from a Parmigiano Reggiano whey processing plant is used to characterize the plant's demand profile (peak load, load factor, and daily operating pattern) within a data-driven energy-monitoring framework. This characterization provides the basis for a seasonal coefficient of performance (SCOP) analysis comparing two heat pump configurations against the plant's current chiller-based baseline: (A) a cooling-only heat pump rejecting heat to ambient air, whose COP depends on climate; and (B) a heat-recovery heat pump condensing against process water at a fixed useful temperature, whose COP depends on process requirements rather than climate. A sensitivity analysis on the key techno-economic parameters completes the comparison. Results show that Scenario A exhibits strong climate-driven seasonality (winter SCOP 6.07 vs. summer SCOP 4.06), while Scenario B maintains a constant, process-driven COP (2.36) throughout the year. Both configurations reduce operating cost and environmental emissions relative to the current chiller, with the heat-recovery configuration achieving the larger annual saving in this case study. The proposed approach illustrates how high-resolution consumption data can support the integration of techno-economic assessment into data-driven decision-support frameworks for industrial energy systems.

Paper Nr: 58
Title:

An Explainable Fuzzy Inverse Surrogate Embedded in a Layered Digital Model for Fiber Extrusion on the FrED Platform

Authors:

Pedro Ponce, José Carlos Saishio, Krysia Abigail Romero, Brian W. Anthony, Russel Bradley and David Ángeles-Rojas

Abstract: Inverse modeling of manufacturing processes often requires computationally intensive analytical formulations, whereas purely data-driven alternatives typically sacrifice transparency and physical interpretability. This trade-off is particularly relevant in fiber extrusion, where estimating the motor voltage needed to achieve a desired filament diameter is essential for future closed-loop control and Digital Twin applications. To address this problem, this work proposes an explainable fuzzy-logic inverse surrogate for the Fiber Extrusion Device (FrED), embedded within a four-layer physics-grounded Digital Model. The proposed architecture supports multiple first-principles representations of the extrusion process and is implemented using an electromechanical inverse model. An alternative thermo-electromechanical formulation is included solely to illustrate the extensibility of the proposed architecture. A Mamdani fuzzy inference system replaces the chained analytical inverse by directly mapping extruder speed and target fiber diameter to the required motor voltage through an interpretable rule base automatically extracted from analytically generated training data using a Winner-Takes-All assignment strategy. A sensitivity analysis across minimum-support thresholds, membership-function shapes, and numbers of membership functions showed that a compact 16-rule model built from four triangular membership functions provides the best balance between accuracy and interpretability. Validation against the analytical inverse and the dynamic ODE model yielded an average motor-voltage error of 4.43% and an average steady-state fiber-diameter error of 2.41%, while preserving complete rule-level transparency. These results demonstrate that the proposed surrogate is an accurate and explainable AI component for physics-grounded Digital Models and provides a practical basis for future closed-loop, sensor-coupled Digital Twins for fiber extrusion.

Paper Nr: 61
Title:

A Data-Driven Generic Simulation Model for Production Workshops with Heterogeneous Process Complexity

Authors:

Victoria Monmagnon, Lucas Braquet, Gilles Viard, Florian Ruas, Esma Yahia and Philippe Veron

Abstract: In the context of Industry 4.0, the ability of companies to rapidly model their production systems is a key driver of competitiveness. However, traditional simulation approaches remain heavily dependent on the structure of the system under study: any change to the workshop or deployment at a new site typically requires developing a new model or significantly reconfiguring the existing one. This article proposes a Data-Driven Generic Simulation Model (DDGSM) designed to overcome this limitation. Its primary scientific contribution lies in the creation of a single model capable of representing a broad range of production systems within a defined process-oriented applicability domain (excluding assembly systems) without altering its architecture. Adaptation to a new industrial environment is achieved solely by updating input data sourced from the ERP system, allowing the company to reuse the same model across different workshops or production sites without modifying the code or simulation logic. This genericity is enabled by an architecture organized into three levels of increasing complexity, balancing deployment flexibility with modelling accuracy: even for complex industrial processes such as heat treatment furnaces. Developed using AnyLogic and driven by an Excel data structure extracted from the ERP, the DDGSM solution has been validated across 43 job-shop scenarios as well as a large-scale industrial case study.

Short Papers
Paper Nr: 38
Title:

Quantifying Domain Randomization and Data Efficiency in a Visual Synthetic Twin for Fiber Extrusion Defect Classification

Authors:

Adán Emiliano González Hernández

Abstract: Synthetic training data is frequently proposed as a remedy for the scarcity of labeled defect images in early-stage manufacturing inspection, yet evidence about which generator design choices matter and in which real-data regimes augmentation pays off remains limited. This paper presents a controlled, 180-run evaluation of a visual synthetic twin, a parametric, physically motivated model of the image formation of a fiber-extrusion inspection scene, for three-class defect classification (glitter contamination, fiber twisting, non-defective). The evaluation comprises three experiments on a fixed unseen real test set: (i) a backbone study crossing six training-data conditions with three convolutional classifiers (YOLOv8m-cls, ResNet50, EfficientNet-B0); (ii) a low-data sweep over real-data fractions from 1% to 100%; and (iii) a component ablation of the generator itself. All comparisons are paired by seed and backbone and assessed with Wilcoxon signed-rank tests, paired t-tests, and effect sizes. Three findings emerge. First, domain randomization is the dominant generator design factor: it improves synthetic-only transfer by +0.188 macro-F1 (p = 0.008, d = 1.34), and ablating it from the generator costs 0.303 macro-F1 (p = 0.014). Second, synthetic-only training collapses to a binary glitter detector (the twisted class reaches F1 = 0.00 for ResNet50 across all seeds), which localizes the sim-to-real gap in the geometric and textural appearance of subtle classes rather than in high-contrast chromatic defects. Third, synthetic augmentation of real data yields at most a marginal benefit confined to the 5–10% real-data regime (d ≈ 0.7, p ≈ 0.06), while a 20% real subset is already statistically indistinguishable from full-real training, establishing an empirical data-saturation floor for this task. This paper contributes a reproducible protocol and quantitative evidence for building synthetic data pipelines for visual inspection models

Paper Nr: 46
Title:

Towards a System Dynamics-Based Tool for Feasibility Assessment of Circular Economy Watermelon Rind Jam Ventures

Authors:

Valentina Ospina Soler, Freddy Peréz, Ernesto Alonso Vega Telles and Camila Josephine Garatuza Hernandez

Abstract: The valorisation of agri-food by-products represents a central strategy within circular economy transitions, yet the feasibility assessment of ventures built around specific waste streams remains dominated by static analytical frameworks that fail to capture the systemic complexity inherent to these business models. This paper positions System Dynamics (SD) as a suitable methodology to address this gap, and proposes an SD-based simulation tool for the integrated technical feasibility assessment of a circular economy venture centred on the production of watermelon rind (Citrullus lanatus) jam in the municipality of Cajeme, Sonora, Mexico. Following a six-phase modelling approach grounded in Forrester [16] and Sterman [25], a stock-and-flow model was constructed and implemented in Stella Architect Lab, representing ten processing stations organised in two parallel production lines with quality-driven material diversion loops. Structural validation through unit-consistency and extreme-condition tests supported the dimensional integrity and logical behaviour of the model. Sensitivity analysis revealed a non-linear amplification effect: a ±25% variation in raw material input rate generates approximately ±40% change in weekly output, underscoring the strategic criticality of supply chain stability. Methodologically, this work joins an emerging current extending SD to the circular bioeconomy domain, and establishes a replicable precedent for simulation-based feasibility assessment of agri-food valorisation ventures.

Paper Nr: 47
Title:

Hybrid Workflow for Offline on-Device RAG: Design Choices Really Matter

Authors:

Yue Cao, Kevin Ducharlet and Houssem Saidi

Abstract: RAG-based solutions are now essential for addressing document-based question answering problems. In the industrial maintenance context, technicians have to work with complex documentation, which is often centralized in computerized maintenance management systems and accessed via a smartphone during maintenance tasks. This industrial context imposes further constraints regarding data security and, at times, even internet connection availability. In such cases, RAG need to operate offline and on-device, with limited capabilities. Only data extraction and preparation can be processed server-side, leading to a hybrid workflow. This constrained RAG problem is gaining prominence in the state of the art, but there is still a real lack of research assessing the impact of the various stages of such workflows and dependencies between those stages. To support this statement, this article highlights the various possible design choices and performance evaluation metrics that can be applied at each stage, emphasizing the complexity of design choices and the need to study them in greater details.

Paper Nr: 55
Title:

An Open Reference Architecture for High-Frequency Non-Destructive Testing Using Unified Namespace and Asset Administration Shell

Authors:

Ander Garcia, Xabier Oregui and Telmo Fernández De Barrena

Abstract: Non-destructive testing (NDT) evaluates the properties of materials, components, and systems without causing damage. Traditional NDT systems are frequently implemented as closed solutions that provide final inspection results while offering limited access to raw measurements, intermediate data, or integration interfaces. This restricts part-level traceability and makes it difficult to incorporate custom analysis algorithms and workflows. This paper proposes an open NDT architecture aligned with Industry~4.0 principles, in which information technology (IT) and operational technology (OT) components are integrated through an event-driven and highly decoupled approach based on open-source technologies. A Unified Namespace (UNS) supports communication among the NDT components, while Asset Administration Shell (AAS)-based digital twins make inspection data and results accessible through dedicated submodels and standardized interfaces. The architecture is validated using an impact-based acoustic inspection system comprising a PLC-controlled conveyor, NFC-based part identification, an impact hammer, an airborne acoustic sensor, high-frequency data acquisition, an edge-deployed Random Forest (RF) classifier, time-series storage, and an AAS server based on Eclipse BaSyx. The experimental validation uses 742 impact-acoustic recordings acquired from 15 physical parts, including ten conforming and five defective parts, sampled at 64~kHz. The implemented system demonstrates the end-to-end integration of part identification, signal acquisition, edge inference, data persistence, and per-part AAS updates, providing a basis for open and traceable NDT applications.

Paper Nr: 56
Title:

Bridging the Experience Gap: Generative AI for Cold-Start in Case-Based Design Support

Authors:

Kokulan Thanabalan, Lukas Malburg and Ralph Bergmann

Abstract: Industrial product design requires decisions that balance technical, economic, environmental, and manufacturing constraints, yet Case-Based Reasoning (CBR) systems often provide limited support when their case bases are sparse or lack sufficiently similar prior experience. This position paper proposes a hybrid decision-support concept that combines CBR with generative artificial intelligence for Design for Manufacturing problems in CAD-based product development to solve the cold start problem. In the proposed approach, validated organizational experience remains the preferred source of recommendations, while Large Language Models generate provisional problem analyses, design alternatives, and solution hypotheses when relevant cases are unavailable. The concept integrates CAD and product data, design constraints, rules, standards, company knowledge, and the four CBR knowledge containers: vocabulary, case base, similarity measure, and solution transformation. A human-centered interface clearly distinguishes retrieved experience from generated suggestions and requires expert review and technical validation before any outcome is retained as a reusable case. The paper further outlines an evaluation plan comparing multiple closed and open-weight models through expert assessment, reproducible prompting, repeated runs, and quality criteria such as feasibility, relevance, constraint compliance, revision effort, and case suitability. Key challenges include hallucinations, incomplete context, data protection, automation bias, and reproducibility.

Paper Nr: 60
Title:

Improvement of the FrED Extrusion Manufacturing Process Using the Earthquake Metaheuristic Optimization

Authors:

José Carlos Saishio, Pedro Ponce, David Ángeles-Rojas, Russel Bradley and Brian W. Anthony

Abstract: Maintaining a stable spool speed response is a critical challenge in the FrED (Fiber Extrusion Device) platform, because variations in DC motor performance directly affect fiber diameter and product quality. PID controllers remain the preferred industrial solution, but their performance depends strongly on the tuning strategy, and metaheuristic optimizers such as Particle Swarm Optimization (PSO) and the Bat Algorithm (BA) can suffer from premature convergence and sensitivity to local minima. This paper experimentally evaluates the Earthquake Algorithm (EA) for PID tuning of the DC motor that regulates the spooling speed of FrED, against PSO and BA under identical conditions. For each algorithm, 24 hyperparameter configurations were sampled and evaluated, and the selected ones optimized the PID gains for four objective functions (ISE, IAE, ITSE and ITAE) under practical constraints on the gains and control effort. PSO reached the lowest cost in all four cases, but varied widely with initialization and hyperparameters, whereas EA stayed within 1.3% of the best PSO cost with markedly greater robustness to both, at the price of more iterations. BA gave intermediate solution quality at the highest computational cost. The twelve controllers were validated on the physical FrED spooler, where the measured responses closely matched the simulations. These findings position the Earthquake Algorithm as a robust alternative for industrial PID tuning and establish a benchmarking framework for metaheuristic optimization in advanced manufacturing control systems.

Paper Nr: 37
Title:

AI-Enhanced KOSU Automation for Reliable Lean Manufacturing in the Automotive Industry: A Case Study

Authors:

Manar Messour and Fayçal Fedouaki

Abstract: In the evolving landscape of the automotive industry, Lean Manufacturing continues to serve as a cornerstone for operational excellence, emphasizing efficiency, quality, and waste elimination. Among its core indicators, KOSU a Lean indicator representing the labor effort required to produce one unit. Plays a critical role in maintaining flow stability and ensuring on-time delivery. However, conventional KOSU systems often rely on manual data entry and delayed feedback, which limits their reliability and responsiveness in dynamic production environments. This paper explores the integration of Artificial Intelligence (AI) and Manufacturing Execution System (MES) technologies to enhance the reliability of KOSU as a Lean performance management tool. The study focuses on the automation of KOSU calculation through the MES, which captures real-time process data (timestamps, test results, operator context) directly from the production line, eliminating human error and latency. Building upon this automated foundation, AI-driven algorithms are employed to predict operator performance fluctuations and potential downtimes by analyzing historical process times, setup delays, and non-conformance rates. Through the case study conducted in an automotive plant, the results demonstrate measurable improvements in data accuracy, responsiveness, and predictive control. The integration of AI models such as Random Forest for operator performance and LSTM for downtime prediction resulted in enhanced KOSU stability, faster reaction to deviations, and reduced unplanned stoppages. This approach showcases how combining MES automation with AI analytics strengthens the fiabilization of Lean Manufacturing, transforming KOSU from a reactive performance indicator into a proactive and predictive decision-support system.

Paper Nr: 39
Title:

The Role of IoT in Enhancing Waste Elimination: A Lean 4.0 Perspective

Authors:

Ayoub Elkhairi and Faycal Fedouaki

Abstract: The integration of Lean Manufacturing and Industry 4.0—often termed Lean 4.0—has generated growing interest in the use of Internet of Things (IoT) technologies to support waste elimination. However, few empirical studies have evaluated how specific IoT applications align with the seven classical types of Lean waste. This paper investigates the role of IoT in enhancing Lean waste elimination through a qualitative multiple-case study of three manufacturing firms actively implementing Lean and IoT systems. Data were collected via semi-structured interviews, IoT system logs, and internal performance reports. Findings reveal that IoT technologies contributed to significant reductions in key waste indicators, including a 15–28% decrease in machine downtime, 20–30% improvement in inventory turnover, and 11–22% reduction in product defects. Thematic analysis highlights three cross-cutting enablers: real-time visibility, data-driven feedback loops, and continuous improvement acceleration. This study contributes to the Lean 4.0 literature by offering evidence-based insights into how IoT supports operational efficiency in real-world settings. It also provides practical recommendations for managers aiming to align digital transformation efforts with Lean principles.

Area 2 - Logistics

Full Papers
Paper Nr: 30
Title:

Impact of the Cooperative Intelligent Transport Systems (C-ITS) on Logistics: A Simulation Case Study of the Port of Marseille

Authors:

Wassim Bouazza, Abdessamad Ait El Cadi and Abdelghani Bekrar

Abstract: With the increasing globalization of trade, ports have become even more strategic, with various heavy vehicles and equipment playing a crucial role in logistics operations. Over recent decades, the integration of pervasive technologies has significantly transformed industries, including transportation and logistics. In particular, new communication technologies are reshaping port operations. Simultaneously, the adoption of advanced vehicle technologies, such as driver assistance and autonomous driving, has grown, offering opportunities to enhance logistics efficiency and reduce greenhouse gas emissions. This article examines the impact of these technologies on road freight logistics in a maritime port, using computational simulations of truck flows around and within the Port of Marseille. Through a comparative study, five different operational modes are tested, ranging from a base mode without any C-ITS to modes incorporating three technologies: Smart Gate, Smart Parking, and Smart Speed Regulation, to assess their impact on efficiency and CO2 reduction. The results show that individually deployed technologies yield only marginal gains, below 1.6% in CO2, and may even degrade traffic fluidity when speed regulation is applied in isolation, whereas their joint deployment reduces CO2 emissions by 7.16% and truck waiting times by 9.44%, an effect larger than the sum of the individual contributions.

Paper Nr: 33
Title:

Spare Parts Classification for Preventive Maintenance: A Fundamental Study and Framework

Authors:

Liwen Zhang, Enzo Pinho Fernandes, Sara Maqrot and Houssem Saidi

Abstract: Preventive Maintenance, Spare Part Management, Classification.In industrial contexts, preventive maintenance aims to prevent equipment failures and reduce potential economic losses. Since spare parts are essential to maintenance interventions and highly diverse, their classification is a key task for supporting decision-makers in defining appropriate management strategies, in line with the human-centric principle of Industry 5.0. In this work, we first review existing classification methods for spare parts and inventory management. We then proposes a framework composed of two applications: 1) a Data Generator, used to create synthetic datasets for testing and benchmarking, and 2) a Classification Application, integrating seven single-criterion methods and nine multi-criteria combinations with reporting functionalities. The framework is evaluated using the M5 dataset and six generated datasets through a set of indicators. The results show its ability to select suitable classification methods and provide high-quality classification outputs.

Paper Nr: 54
Title:

From Nominal to Usable Inventory: Provenance-Gated Decision Support for Telecommunications Logistics under Disruption

Authors:

Lakshmanaprakash Murugesan

Abstract: Telecommunications logistics depends not only on the quantity and location of stock, but also on whether inventory is identifiable, reconciled, trusted, and eligible for allocation within a disruption-response window. Existing forecasting, shipment-visibility, inventory-optimization, and traceability approaches are commonly evaluated as separate capabilities, leaving limited evidence on how uncertain inbound information becomes decision-ready supply. This paper develops a provenance-gated decision-support model in which usable inventory acts as the connecting state between advance ship notice confidence, receipt evidence, inventory positioning, and governed allocation. The study uses a two-layer evaluation. First, de-identified operational telecommunications planning data compare an XGBoost forecast with a recent- history baseline. Second, a controlled Monte Carlo simulation evaluates architecture-level effects under a compound hurricane scenario using 100 paired replications and common random numbers. Forecasting wMAPE improves from 14.54% to 11.78% overall and from 14.26% to 10.90% outside the disruption window, but the AI model underperforms during the hurricane period. In the controlled simulation, provenance-gated decision support reduces false availability from 4.8% to 1.1%, improves fill rate from 91.3% to 96.0%, and shortens modeled recovery time from 25.9 to 15.4 days. The results indicate that prediction is most useful when coupled with trusted inventory-state logic; field validation remains necessary.

Short Papers
Paper Nr: 18
Title:

Consistency-Aware Online Decision-Making: Competitive Analysis with Trajectory Stability

Authors:

Christine Markarian, Claude Fachkha and Haris M. Khalid

Abstract: Many modern production and computing systems, including cloud platforms, large-scale infrastructures, and Artificial Intelligence (AI)-driven systems, must make decisions sequentially as requests arrive, without knowledge of future inputs. Algorithms operating in this setting are studied as online algorithms. Their performance is evaluated relative to an ideal offline algorithm with full knowledge of the input sequence. Competitive analysis provides the classical framework for such comparisons and quantifies efficiency under uncertainty. While competitive analysis captures efficiency, it does not account for how decisions evolve over time. In practical systems, frequent decision changes may introduce instability and operational overhead, even when efficiency remains unchanged. Consequently, two online algorithms may achieve identical competitive guarantees while exhibiting fundamentally different levels of trajectory stability. This work introduces Consistency-Aware Competitive Analysis, a framework that extends competitive analysis by incorporating trajectory stability as a formal performance dimension. Consistency is defined as the cumulative variation of decisions over time. The proposed Consistency Competitive Ratio (CCR) evaluates algorithms jointly with respect to efficiency and cumulative variation relative to an optimal offline solution. The framework is studied through online scheduling on identical parallel machines. The analysis shows that algorithms with optimal classical guarantees may incur unbounded cumulative variation. It establishes bicriteria guarantees that simultaneously bound makespan and cumulative variation.The analysis further reveals a fundamental limitation: no online algorithm can simultaneously guarantee optimal efficiency and bounded cumulative variation on all input sequences.

Paper Nr: 24
Title:

Predicting Dimensional Weight in Logistics: A Comparative Study of Traditional and Modern Methods

Authors:

Timuçin Çınar, Fatma Kutlu Gündoğdu, Melisa Şimşek, Selçuk Cebi, Hamide Uslu and Ersin Şengül

Abstract: Operational efficiency in logistics management increasingly depends on the precision of demand forecasting, particularly in predicting package volumes at distribution, transfer, and collection facilities. Although previous studies have examined shipping quantities, accurately estimating dimensional weight poses a more complex task, essential for enhancing workforce planning, vehicle load capacity, and delivery efficiency. This study performs a comparison of traditional and recent forecasting methods such as Holt-Winters, Prophet, and Multi-Layer Perceptron (MLP) utilizing historical shipment data that reflects variations in cargo density. The paper comprehensively examines current forecasting strategies, emphasizing the shift from traditional time-series models to sophisticated machine learning and deep learning algorithms adept at high-dimensional logistical data. Driven by the operational disturbances resulting from erroneous volume estimations, the study assesses the prediction efficacy of each model utilizing error metrics like MAPE. This study, utilizing real-world data, revealed that the efficacy of each approach differed by location. The results suggest that forecasting outcomes are contingent upon the continuity of datasets, operational variables, and seasonal cycles. The findings aim to identify the most effective strategy for dimensional weight forecasting, supporting more reliable and cost-efficient logistics operations. Recommendations for future research and industry practice are also discussed.

Paper Nr: 40
Title:

Health Logistics Optimization in Low- and Middle-Income Countries: A Scoping Review of Graph-Based Network Analysis and Vehicle Routing Approaches

Authors:

Saan-nonnan Olivier Dabire, Boureima Zerbo and Désiré Guel

Abstract: Healthcare supply chains in low- and middle-income countries (LMICs) face persistent challenges stemming from fragile infrastructure, fragmented information systems, and weak coor- dination, all of which compromise equitable care delivery. Whilst network diagnostics and logistics optimisation are conventionally addressed in isolation, this position paper advocates their inte- gration within a unified decision-support framework. Drawing on a PRISMA-ScR-guided scoping review of 39 studies (2017–2024) from 200 initial records across six databases, the proposed frame- work combines graph-theoretic centrality analysis for structural vulnerability identification with vehicle routing problem (VRP) optimisation under time-window, cold-chain, and fleet-capacity constraints. To move beyond conceptual theorisation, the framework is instantiated through an illustrative application to Burkina Faso’s health supply chain, supported by a mitigation matrix addressing key implementation barriers. It is argued that hybrid architectures coupling compu- tational optimisation with governance mechanisms offer the most promising foundation for re- silient, sustainable health logistics in resource-constrained settings. Grounded in Burkina Faso’s institutional context as a transferable model for comparable LMICs, and positioned within the Operations Research for Development (OR4D) agenda, the framework remains conceptual in its empirical scope; adaptive learning and field validation constitute key priorities for future research.

Paper Nr: 51
Title:

An Explainable Hybrid AI Framework for Interoperable Digital Trade Compliance Intelligence

Authors:

Adewale Ogabi, Michael Short and Geetika Aggarwal

Abstract: The rapid digitalisation of international trade, together with emerging regulatory frameworks governing carbon disclosure, rules-of-origin verification, and digital product passports, has created a growing need for intelligent and interoperable compliance assessment. Existing digital trade infrastructures primarily support document exchange and interoperability but provide limited analytical verification of heterogeneous compliance records. This research proposes a hybrid explainable AI framework that formalises digital trade compliance as an AI-assisted pre-screening problem over an ICC-informed canonical analytical representation. The framework integrates deterministic compliance validation, unsupervised anomaly detection, supervised risk classification, and hybrid evidence fusion. A prototype is evaluated using 10,000 synthetically generated aluminium trade declarations containing predefined compliance anomalies. The hybrid framework achieves precision of 0.864, recall of 0.886, F1-score of 0.875, ROC-AUC of 0.967, and PR-AUC of 0.936, while supporting transparent risk prioritisation across heterogeneous anomaly classes. The principal contribution is a Compliance Intelligence Layer that separates trade-data interoperability from analytical compliance verification and combines heterogeneous evidence for explainable, human-in-the-loop pre-screening.

Paper Nr: 59
Title:

Improving Supply Chain Operational Performance through Technology Integration and Lean-Agile Practices

Authors:

Meryem Bouraoud, Hafida Benseddik and Salima El Omari

Abstract: This study examines and analyzes the links identified in the literature between information technology (IT) and two supply chain strategies that share identical mechanisms of action: the “Lean” supply chain (LSC) and the “Agile” supply chain (ASC). The objective is to identify and understand the role of IT based on its typology (ERP, blockchain, AI, IoT, etc.) and its various dimensions (technological, organizational, strategic) within these strategies, as well as its impact on the operational performance of companies. The methodology is based on a systematic literature review, which aids in the identification, selection, processing, and analysis of relevant studies. Complementing existing research, this study evaluates, analyzes, and examines the impact of information technology on hybrid Lean-Agile (Leagile) models, which have often been underrepresented in literature. Furthermore, it draws on Dynamic Capabilities Theory (DCT) and the Resource-Based View (RBV) to explain how companies mobilize and reconfigure their information technologies to enhance their competitive advantage. The intersection of these theoretical frameworks provides insight into performance differences among companies based on their ability to adapt and integrate information technology into their supply chains. This research highlights the need for a contextualized and nuanced approach to conceptualizing information technology and offers insights for future academic and managerial research.

Paper Nr: 27
Title:

HYDRA-Net: Hybrid Deep Routing and Arrangement Network

Authors:

Dhawal Thakkar, Dhanish Ahammed M. V. and Balamurugan Palaniappan

Abstract: This paper addresses the Integrated Capacitated Vehicle Routing Problem with Time Windows and 3D Loading (3L-CVRPTW), which generalizes the classical VRPTW by incorporating three-dimensional packing constraints. To solve this challenging problem, we introduce a novel deep learning framework, HYDRA-Net (HYbrid Deep Routing and Arrangement Network), which unifies two specialized transformer-based architectures via an integrated inference logic. The system decomposes the problem into TRANSIT-Net (Transformer Vehicle Routing Architecture for Networks, Scalable and Intelligent) for routing, and STACKNet (Spatial Transformer Architecture for Container Kinematics) for 3D loading. The routing model predicts site sequences, time bucket allocations, and vehicle assignments using a multimodal cross-attention mechanism, while the loading model ensures geometric feasibility through a constraint-aware autoregressive process while trying to minimize unloading effort. The framework is trained in a supervised self-improving fashion on synthetically generated instances, using high-quality baseline solutions as training labels. These labels are generated by a custom hybrid heuristic combining K-Means clustering, Ant Colony Optimization, and a constructive grid-based packing algorithm with surface support constraints. Experiments demonstrate that our integrated approach delivers strong performance, effectively minimizing operational costs while guaranteeing physical executability. The model prevents the operational failures common in non-integrated approaches. This work bridges the gap between theoretical routing and practical logistics, offering a scalable, adaptive framework for real-world urban delivery scenarios where physical constraints are as critical as temporal ones.