The paper industry is highly energy intensive and exposed to volatile energy markets and strict decarbonisation targets. Digital twins can couple operational data with engineering and simulation models for energy and CO2eq management, yet many solutions remain domain-specific. This paper summarises FOREST (Framework for Resource, Energy, and Sustainability Treatment) as a modular digital twin framework that integrates plant data, Asset Administration Shell (AAS)-based models, and executable simulations to provide consistent energy and material flow views, time- and product-resolved CO2eq indicators, and scenario evaluations. Based on an industrial tissue implementation, three domain-independent architectural patterns are derived. flow and state tracking, asset-centric semantics, and co-execution of models and services. These patterns are conceptually mapped to electric traction motor manufacturing and End-of-Use (EoU) testing within industrial data spaces, illustrating how plant-level optimisation logic can be extended towards lifecycle-oriented decisions on value-retention strategies.
Manufacturing firms face growing exposure to volatile and uncertain global markets. Particularly in make-to-order (MTO) environments involving complex, large-scale machinery and equipment, long project lead times amplify the impact of external cost fluctuations on competitiveness and profitability. To date, limited attention has been given to how firms can systematically integrate dynamic external market factors into early-stage project costing. This study addresses this gap through an in-depth single case within the Swiss mechanical engineering industry. Using a Design Science Research (DSR) approach, a conceptual frame-work and cost prediction tool were developed to capture the effects of dynamically changing cost factors - such as wages, exchange rates, and material prices - on distinct cost components, including in-house pro-duction, external devices, and service activities. The framework was operationalized in a dynamic cost model employing exponential smoothing (Holt-Winters) time-series forecasting based on monthly historical data. The findings demonstrate that exchange-rate fluctuations and regional cost differentials can materially alter projected project margins and reshape the relative attractiveness of alternative production locations and as-sociated supply chain configurations. This study contributes to the literature on MTO costing and adaptive supply chain configuration under uncertainty by showing how external market volatility can be operational-ized within early-stage decision processes. For practitioners, this tool provides early cost transparency and supports the evaluation of alternative supply chain configurations and pricing decisions.
Deviation management (DM) is critical for production stability and continuous improvement, yet much shop-floor knowledge remains unstructured, limiting effective decision support. This paper proposes a Knowledge-Based Decision Support System (KB-DSS) for DM. Based on a systematic literature review (SLR), a four-layer framework comprising data, knowledge, application, and user interface layers is designed. The framework integrates heterogeneous deviation-related information into structured, actionable knowledge and links it to real-time deviation events. By leveraging data-science techniques and large language models, the system supports knowledge retrieval, reasoning, and acquisition across the DM workflow. The KB-DSS is implemented and evaluated in multiple industrial environments, demonstrating its scalability and practical applicability for knowledge-driven DM in manufacturing.
In recent years, research in social robotics has gained popularity in several application fields, most prominently healthcare. Enhanced human-robot interaction provides added value in elderly care, where social capabilities can improve the quality of life. Likewise, the topic has also become relevant in other fields, such as manufacturing. A deployed robot's social skills can increase the perceived comfort of collaborative tasks, such as resolving machine malfunctions or training new workers, improving the overall productivity and robot acceptance in modern factories. However, their validation through digital approaches proves to be unarguably difficult. Without a formal tool to encode, store, and retrieve social knowledge about scenarios of industrial human-robot interaction, there is no way for a computer to tell how a complex behavior model with a multitude of social abilities performs in terms of human trust, user acceptance, or naturalness of an interaction. Due to the imprecise nature of social facts and knowledge, encoding this information is a nontrivial task. Mathematical models based on fuzzy logic and fuzzy sets have the potential to express a social robot's behavior and cognition. Therefore, the authors propose a prospective approach with corresponding design goals towards an explainable social knowledge base through semantic formalization and fuzzy techniques as an initial step to provide said tool for social robot validation as an explainable rule-based system. An extensive and reproducible systematic literature review on applications of fuzzy techniques in social robotics supports the argumentation. The used review scheme is PRISMA2020.
The integration of Lean Six Sigma (LSS) principles with Industry 4.0 technologies offers an effective way to improve operational performance in modern manufacturing environments. This case study presents the systematic application of LSS within an Industry 4.0 manufacturing laboratory, employing the DMAIC (Define, Measure, Analyze, Improve, Control) framework to identify and eliminate inefficiencies. In a context dominated by automation, cyber-physical systems, and data-driven decision making, the study addresses persistent challenges including inaccurate sensor readings, limited user knowledge of advanced systems, and suboptimal material storage configurations. Through a combination of statistical analysis, root-cause identification, and iterative process redesign, the intervention resulted in measurable improvements in process reliability, user experience, and overall workflow efficiency. The findings highlight how LSS can serve as an effective bridge between traditional continuous-improvement methodologies and digitalized manufacturing operations. Moreover, the outcomes position the integration of LSS as a foundational enabler for Industry 5.0, where human-centricity, resilience, and sustainability will increasingly shape manufacturing system design.
Configuring pick-and-place robots for machine-tending tasks typically requires several days of expert programming and cell setup. This paper presents a digital-twin approach that enables non-experts to program real robots through extended-reality (XR) interaction. Within an XR environment users position the robot end effector, issue spoken commands for perception and motion planning, and automatically generate executable behaviour-tree programs deployable on a physical robot. While functional execution is demonstrated, future work will focus on quantitative accuracy validation. In a user study, novice participants completed a full six-object pick-and-place sequence in under four minutes. Compared with integrator-based workflows, the approach reduces deployment time by an order of magnitude and supports agile, human-centred automation in smart manufacturing.
Inventory accuracy is a key element for maintaining reliable and efficient logistics operations. In this context, this study analyzes record accuracy and proposes innovative digital solutions. This study addresses the problem of inventory record accuracy (IRA) in white-goods warehouses, where a huge discrepancy was detected between the physical inventory and the system records, resulting in an accuracy level of 89%, below the industry benchmark of 98%. This situation generates annual economic losses of USD 129,888, rework, and low reliability in logistics decision-making. An integrated improvement model is proposed, based on three tools: Pick-to-Light, aimed at reducing picking errors through visual guidance; RFID with overload alerts, designed for real-time physical validation and traceability; and a digital Poka-Yoke, implemented to automatically block incorrect or incomplete records. The study demonstrates that the integration of digital control and traceability tools represents an innovation applicable to logistics management, combining Lean technologies with intelligent validation systems to address inventory accuracy from a preventive and automated perspective. This approach goes beyond traditional physical audits, proposing a layered control model that ensures real-time coherence between physical and digital inventory. Its modular and adaptable structure allows the proposal to be replicated in other industrial sectors or logistics contexts, where traceability and inventory accuracy are critical factors for supply chain efficiency. Overall, the study contributes a validated and replicable methodological framework that advances knowledge in digital logistics management and continuous improvement. Validation was carried out through computational simulation, comparing the current and improved scenarios, revealing an increase in inventory accuracy from 89% to 98%, a significant decrease in picking, dispatch, and operational errors, and annual savings exceeding USD 75,000. The application of digital and control tools under a Lean-logistics approach constitutes an effective alternative to strengthen traceability, security, and inventory reliability in the white-goods sector.
Scientific credibility is essential for both researchers and practitioners alike. Irreproducible empirical research or questionable citation practices lead to problems regarding transparency or simply hinder scientific discourse and knowledge gain. Especially regarding frequently used performance indicators, such as production cost or lead time, industry and academics need credible data alike for their strategic orientation, like investment decisions or the derivation of research gaps and hypothesis. This paper examines citation patterns of fundamental production and assembly indicators. In particular, the key performance indicators of assembly cost and assembly time are analyzed twofold, based on a highly cited German textbook on assembly. First, a citation analysis that looks back at past research is performed to show the origins of the assembly performance indicators. Second, a citation content analysis looking forward is carried out on literature citing the German assembly textbook to provide an overview of the relevance of the examined performance indicators. Lastly, the findings are discussed and implications for scientific credibility in assembly and production research are derived. The results demonstrate that the origins of the exemplary assembly performance indicators are questionable and difficult to determine, as older literature becomes distorted, is inaccessible or fails to give credit to previous work. Furthermore, the second literature review reveals that, despite the indistinct origins of the assembly performance indicators, they remain highly cited. With the indicators being used for motivating many authors' works or as state of the art, an adverse impact on scientific credibility is identifiable.
The concept of cloud manufacturing leads to a significant paradigm change within the industrial context: Many technical hurdles are being mastered, mostly based on digitalization and networking of production sites. Furthermore, production planning and control must evolve in the management dimension in order to address new kinds of concurrent yet independent production processes. However, centralized or hierarchical production planning is inadequate for the different participating entities with individual, diverging goals that interact within and across production sites. In this work, we propose that agent-based modeling more accurately reflects entity behavior intricacies and interrelated interactions within a dynamic yet structured environment. This environment comprises individual production sites with a range of capabilities based on their internal structure and transportation logistics between those sites. Production agents utilize these networked sites for their own production while protecting their intellectual property from other agents by limiting information exchange with other agents, thereby forming a multi-agent system. Hence, the agent-based modeling technique can facilitate modeling and predicting agent performance. One major aspect of this modeling technique in the context of cloud manufacturing is the implicitly defined (in-)flexibility of an agent, which greatly influences its performance within the simulated dynamic environment. This implicit flexibility within agent behavior, as well as its boundaries, are analyzed and quantified within this work.
Accurate, reliable and timely data are the foundation for decision-making in production planning and control (PPC). Relying on an ever-growing number of heterogeneous and interconnected data sources and increasing internal and external uncertainties, software systems are required to gather and process data utilized in decision-making processes. At the same time, data quality is often viewed as given in production research while interconnected production facilities, cyber-physical systems, workers and software systems form a complex network for data exchange and propagation. This paper identifies and categorizes data quality deficiencies relevant to PPC tasks across strategic, tactical and operational planning horizons in order to identify the specific requirements of the tasks. To achieve this, a structured literature review was conducted to study the influence of data quality on the performance of production systems. First, the existing literature was analyzed to determine data quality deficiencies that occur in PPC. Based on the findings the cause-and-effect relationships between data quality and the propagation of these deficiencies through data processing, management, decision-making, and production performance were examined. The results indicate that data quality deficiencies lead to manifold issues, from limiting the effectiveness of PPC tasks to the failure of automated PPC solutions in production after successful prototyping.
The manufacturing industry is currently undergoing a fundamental shift away from linear economic models toward a circular economy. A fundamental principle of the upgrade circular economy lies in the extension of product lifecycle, which extends beyond the mere closure of material cycles through recycling. Thus, a variety of measures may be employed to extend the product lifecycle, including reuse, refurbishment, or product upgrades through technical changes. The upgrade circular economy focuses on the prolongation of the product lifecycle through value-enhancing upgrades in order to preserve resources, increase customer satisfaction and foster sustainability. However, enhancing existing products without modifying their underlaying architectural structure results in reduced flexibility when it comes to incorporating technical modifications and novel features. Thus, the upgradability may be constrained through the existing product architecture, due to the lack of compatibility or outdated interfaces. Companies need to consider both current functional capabilities and potential future adaptability when managing the product lifecycle. This necessitates the evaluation of the realization effort, whether novel features can still be integrated into the existing product architecture. In this regard, it is necessary to consider both the development effort for the upgrade and the reassembly effort for the implementation into the existing product. This paper presents a holistic overview of criteria to evaluate the realization effort of product upgrades in context of the upgrade circular economy. A methodology is developed to evaluate the realization effort of upgrades for existing products, examining the interdependencies between the scope of the technical change and criteria for development and reassembly effort.
Fluctuations in the volume of returned beverage containers can lead to challenges in resource allocation and workforce planning, particularly for smaller collection and counting centers. Forecasting of incoming beverage container quantities is therefore of interest. Accurate predictions enable more efficient distribution of personnel and operational resources across collection and counting facilities, thereby improving logistical performance and cost-effectiveness within the overall deposit-return system. This study presents the application and evaluation of data-driven forecasting models designed to predict the quantity of returned beverage containers differentiated by material type across five counting centers in Germany on a horizon of four weeks and one year. The modeling approach integrates both traditional time series techniques and machine learning algorithms. In addition to the data from the five counting centers, external variables such as meteorological conditions and seasonal and holiday information are incorporated into the analysis. To appropriately account for temporal dependencies during model assessment, a rolling cross-validation scheme is employed. Model performance is quantified using the Mean Squared Error metric. Our findings indicate that DHR (Dynamic Harmonic Regression) and TBATS (Trigonometric Box-Cox transform, ARMA errors, Trend and Seasonal components) models yield promising results for forecasting the return volumes of the different beverage container materials. Their effectiveness in our use-case can be attributed to their explicit capability to capture complex seasonal patterns inherent in the data. The incorporation of external explanatory variables has mixed outcomes on model performance.
Although several concepts have emerged over time that support the circular economy (CE) either directly or indirectly, it is rare to find any studies that try to use them as an integrated approach or even to complement each other towards the final goal. Even though there are some overlaps of these concepts, which address the end-of-life (EoL) phase of products, there is a notable lack of studies that attempt to integrate each of these concepts by considering the stakeholder concerns and the ground-level realities promoting the circular economy. However, synergies of product EoL concepts could contribute toward the CE to a greater extent. Therefore, the present study examines how reverse logistics (RL), urban mining (UM), and closing the loop of the supply chain can be integrated to support the circular economy. The study proposes an integrated decision support framework, identifies different options and avenues available towards the recovery process, and proposes different decisions at various levels of the EoL stage of the product towards each stakeholder. Framework effectiveness is validated through the Material Circularity Indicator analysis of the washing machine recovery process, achieving circularity higher than isolated implementations.
The increasing complexity of automated production systems presents challenges for both industrial and educational environments. Testing, validating and optimizing control strategies directly on physical equipment is often associated with safety risks, limited availability of systems and considerable effort in terms of time and resources. In academic contexts, these constraints additionally reduce opportunities for hands-on learning, as training equipment is frequently limited and requires close supervision. To address these limitations, this research presents the development of a digital model of an educational Fischertechnik production plant controlled by programmable logic controllers (PLCs). The real plant was systematically analyzed and all relevant mechanical, electrical and control parameters were documented. Based on this, a detailed 3D representation of the system was created and integrated into Tecnomatix Plant Simulation by Siemens, where the spatial configuration, process logic and material flow behavior were implemented to reproduce realistic operational sequences and system dynamics. The digital model enables virtual commissioning and supports experimentation and modification of control strategies in a risk-free virtual environment. It increases flexibility in teaching scenarios, reduces dependency on physical hardware and allows users to explore and understand automation processes and production system behavior more efficiently.
This paper presents the development and evaluation of an AI-enhanced smart glove system designed to enable scannerless execution of pick and stow operations in industrial warehouse environments. The core innovation lies in the combination of a wearable device that captures precise hand movements and spatial positions via an indoor localization system, and a novel AI-based method for clustering and combining pick and stow orders. This integration allows an automated detection and confirmation of material handling operations based on walking distance optimized pick and stow order packages without manual barcode scanning or user input. Real-time motion and sensor data from the glove is processed using machine learning algorithms to identify pick and stow activities and detect errors based on positional discrepancies. Detected anomalies are immediately communicated to the operator through acoustic feedback. In parallel, incoming pick and stow orders from ERP systems are dynamically clustered and transformed into optimized task packages. These are autonomously allocated to the nearest available operator, based on spatial proximity and system workload, enabling self-directed and efficient task execution. The system is tested and validated within Werk150, the research and learning factory at Reutlingen University, which replicates realistic assembly and intralogistics processes for variant-rich low-volume production. The work has been conducted as part of the collaborative research project 'ALISA AI- and Localization-Based Order Picking System with Intelligent, Scannerless Work Glove' involving leading small and medium-sized enterprises (SMEs) and academic institutions specializing in logistics and resilient systems. The project focused on integrating intelligent wearables with AI-based decision-making to increase responsiveness and adaptability in intralogistics. The results achieved confirm the feasibility and effectiveness of the proposed approach. Replacing conventional scanning systems with intelligent, AI-supported wearables leads to measurable improvements in process efficiency, error reduction, and operational flexibility in dynamic warehouse environments.
Although Deep Reinforcement Learning (DRL) offers adaptive control for manufacturing processes, training agents within high-fidelity commercial Discrete Event Simulation (DES) tools is hindered by the temporal mismatch between the continuous-time event logic of DES and the discrete-step nature of RL. Current loose-coupling methods (e.g., socket communication) frequently introduce latency and state drift, violating the Markov assumption necessary for stable training. This study introduces a strictly blocking, in-memory synchronization architecture that aligns DES execution with Generalized Semi-Markov Decision Process (GSMDP) decision epochs. Unlike asynchronous approaches, our framework freezes the simulation clock at decision points to ensure near-zero-latency state observations. We validated the architecture using a Dual-Resource Constrained Job Shop Scheduling Problem (DRC-JSSP) involving sequence-dependent setup times. Evaluations on this test bed demonstrate that the agent can learn near-optimal policies with a deviation of only 2.24% from the optimal CP-SAT solver, even under complex constraints where synchronization errors would otherwise lead to infeasible schedules.
Unstructured meetings are a major source of efficiency-related losses in organizations. To address this, a role-based, multi-agent meeting support framework powered by large language models (LLMs) is introduced to enhance structured preparation for cross-functional collaboration. In this framework, each agent simulates a domain-specific role (e.g., logistics, assembly, machining), contributing specialized knowledge to generate targeted agendas and prepare contextual materials aligned with standard operating procedures (SOPs). The approach was evaluated in a learning factory environment using a simulated Kaizen case study, demonstrating its ability to generate goal-driven agendas. Initial results show that well-structured agenda creation is possible, and that the framework can support goal-oriented, cross-functional meetings by generating actionable agendas and providing relevant, explainable information for stakeholder selection. By improving agenda relevance and reducing coordination overhead, this framework provides a scalable method to integrate intelligent meeting support into continuous manufacturing improvement processes, offering a novel application of LLMs to support cross-functional continuous improvement initiatives in manufacturing.
The Digital Product Passport (DPP) is emerging as a key instrument in the European Union's transition toward a transparent, traceable, and circular economy. Building on the regulatory frameworks of the Battery Regulation, the Ecodesign for Sustainable Products Regulation (ESPR), and the draft standards of the DIN EN 18216 - 18246 series, the DPP establishes a service-oriented digital infrastructure for standardized information exchange across product life cycles. Its implementation requires a complex ecosystem of interacting stakeholders, including Economic Operators (EO), Service Providers (SP), and downstream actors each governed by distinct access and modification rights. Within this ecosystem, the management of granular data interactions poses major technical and organizational challenges. A multi-layered architecture based on Role-Based Access Control (RBAC) is essential to ensure data integrity, interoperability, and security across distributed systems. The interaction between actors and central DPP system functions introduces potential failure points, particularly where access permissions or sequencing logic are not clearly defined in the legislation. Moreover, uncertainties within the standardization process, combined with the limited timeframe for mandatory regulatory implementation, increase the complexity of establishing a compliant and reliable DPP infrastructure. In this context, the DPP ecosystem illustrates the extent to which regulatory requirements, technical system design, and stakeholder coordination must align to create a secure and interoperable foundation for product governance in Europe.
Autonomous Mobile Robots (AMR) increasingly operate in complex industrial environments, where reliable detection and classification of near-ground objects is critical for safe and efficient navigation. Near-ground obstacles such as cables, cables ducts, door thresholds, pallets and packaging materials pose significant challenges due to their low height, visual ambiguity and sensitivity to lighting, shadows and floor conditions. Existing perception systems often overlook these objects and fail under dynamic, industrial conditions, revealing an existing gap in current AMR perception capabilities. Therefore, this paper presents a structured requirements analysis for near-ground object detection in AMR. By integrating operational constraints, environmental influences and robot motion characteristics, this paper identifies the technical and functional requirements necessary for robust perception for near-ground objects at floor level. The resulting requirement framework provides a foundation for researching and developing improved detection and classification methods, enabling more reliable navigation in real-world industrial settings.
The rapid growth of e-commerce and the shift in consumer behavior accelerated by COVID-19 have intensified demand for ultra-fast delivery, driving the emergence of quick commerce (Q-commerce). This model relies on a two-echelon structure in which urban Micro Fulfillment Centers (MFCs) receive consolidated shipments from Distribution Centers (DCs) and dispatch orders to nearby customers. While enabling rapid fulfillment, Q-commerce brings challenges in designing and operating a multi-echelon network under uncertain demand, limited storage capacity, and strict delivery-time constraints. This study addresses an integrated Location-Inventory Problem (LIP) for a Q-commerce network composed of DCs, MFCs, and customer zones. The model jointly determines MFC locations, customer-MFC assignment decisions, and product-specific base-stock levels under a periodic-review policy, while allowing reactive lateral transshipments among MFCs to rebalance their inventories in order to reduce stockouts. Demand uncertainty is captured through a scenario-based Mixed Integer Linear Programming (MILP) model with non-anticipativity constraints and a maximum-service-radius restriction reflecting instant-delivery requirements. To efficiently solve the large-scale MILP, we use an enhanced Benders decomposition approach that integrates Pareto-optimal cuts to accelerate its convergence. Numerical experiments validate the proposed model and demonstrate the efficiency of the approach.
