Remanufacturing processes are characterized by high uncertainty due to the variable conditions of returned cores, which makes automation challenging and necessitates considerable process flexibility. Industry 4.0 methods are often proposed to mitigate this uncertainty, yet real-world demonstrations that validate their effectiveness remain limited. This study addresses this research gap by presenting a flexible, digital-twin driven object transfer system implemented in a remanufacturing demonstration environment. The system under consideration involves an autonomous mobile robot that docks at multiple stationary transfer points to transfer unique starter motor cores without the use of load carriers. Since the object transfer process is probabilistic, virtual models are employed in a physics-simulated environment to predict object-specific pre-transfer states, defined as the state an object before the transfer is executed. The predictive capabilities of the digital twins are evaluated through an extensive experimental study, involving a series of physical and virtual experiments conducted on 37 unique starter motor cores.The study includes calibration experiments to tune the virtual models, followed by large-scale virtual experiments to estimate the probability of successful transfer for a fixed set of pre-transfer states. A custom method is applied to determine the most promising pre-transfer state for each starter motor core. Final validation results highlight the effectiveness of the approach and indicate that increased modeling efforts reveal inherent limitations in the predictive accuracy of the virtual models. Sources of error, including mass distribution approximations and simulation inaccuracies, are discussed, and directions for future improvements are outlined.

In the context of circular production, factories must cope with uncertainty arising from the reuse of components with varying availability, quality, and timing. This creates new requirements for intralogistic systems that are both highly flexible and able to adapt to local shifts in transport demand or unforeseen events. We present a novel modular intralogistic system designed to meet these challenges, where autonomous mobile robots can mount self-contained modules to provide on-demand reconfiguration of material-handling capabilities. To ensure semantic interoperability within the circular factory, we further introduce the accompanying ontologies that formalize key concepts and support knowledge-driven decision-making. 

New economic systems are needed to decouple resource consumption from wealth. The linear economic approach of “take-make-use-dispose” is not a recipe for success in the long term. Circular production offers a solution to this problem. Our Collaborative Research Center 1574 aims to enable integrated linear and circular production on an industrial scale. To this end, the Collaborative Research Center 1574 investigates how used products and their multiple generations can achieve the vision of the perpetual product. This research involves multiple scientific questions related to production technology, product development and materials technology, ergonomics, robotics, computer science, and knowledge modeling. The Collaborative Research Center 1574 involves eighteen subprojects. Each one studies a dimension of these multiple scientific questions. One of these subprojects, called INF, aims to provide an infrastructure and teach the other subprojects’ researchers how to operate and integrate all the data they produced into a unified knowledge graph. This paper describes the roadmap of the INF subproject, including the ongoing work, future steps, and vision of the INF subproject.

Large Language Models (LLMs) show strong reasoning abilities but rely on internalized knowledge that is often insufficient, outdated, or incorrect when trying to answer a question that requires specific domain knowledge. Knowledge Graphs (KGs) provide structured external knowledge, yet their complexity and multi-hop reasoning requirements make integration challenging. We present ARK-V1, a simple KG-agent that iteratively explores graphs to answer natural language queries. We evaluate several not fine-tuned state-of-the art LLMs as backbones for ARK-V1 on the CoLoTa dataset, which requires both KG-based and commonsense reasoning over long-tail entities. ARK-V1 achieves substantially higher conditional accuracies than Chain-of-Thought baselines, and larger backbone models show a clear trend toward better coverage, correctness, and stability.

To mitigate the effects of global warming, governments worldwide have set targets for reducing greenhouse gas emissions. Achieving these targets necessitates a socio-ecological transformation in the industrial sector, shifting from the linear "take, make, waste" model to a circular economy. Remanufacturing, one of the end-of-life strategies, is often considered to be the most effective strategy for saving materials and energy (Nasr and Thurston 2006). However, economic barriers at both the supply chain and factory levels hinder the broader implementation of remanufacturing across various industries. The primary challenge is the high level of uncertainty associated with returned products, requiring significant process flexibility that current automation approaches cannot accommodate. Industry 4.0 methods are frequently proposed as a solution to reduce this uncertainty in remanufacturing processes. Nonetheless, there is a notable scarcity of implementations that effectively demonstrate these benefits, highlighting a critical research gap.

In this thesis, we contribute to closing this gap by demonstrating a real-world application of a flexible, digital twin-driven object transfer system. Our contributions are four-fold. First, we propose a non-hierarchical Digital Twin architecture.Second, we develop a formal description model for a generalised object transfer system. Third, we implement both the architecture and the model in a real-world cyber-physical system for object transfer within a remanufacturing demonstration environment. Finally, we conduct an application study using the proposed system, where digital twins are employed to forecast the most promising pre-transfer states for starter motor cores that have never been physically transferred.

The proposed Digital Twin architecture is characterised by a non-hierarchical co-existence of both physical and logical assets (including virtual models) that form a cyber-physical system through the interconnection of their virtual representations.The architecture emphasises decentralised interactions and ensures semantics through the integration of a knowledge graph.The description model of a generalised object transfer system centers around a description of the object state. We describe the requirements for the formation process of the an object transfer system and the process itself through a state transition function. The process is driven by continuous evaluation of the object state. We introduce several important sets of states and demonstrate how the approach can be utilized in different object transfer scenarios.

We implemented an object transfer system based on a vision-controlled autonomous mobile robot and stationary transfer units to realise the flexible transfer of a wide range of objects without the need for a load handling device. The physical assets are also virtually modeled in a physics-engine-based simulation environment.

In our application study, we conduct an extensive set of both real-world and virtual experiments for transferring a 37 unique starter motor cores. The virtual models were initially calibrated using a small set of real-world experiments, resulting in a F0.5 score of 0.84 when comparing the binary predictions between the virtual and physical systems. Subsequently, the calibrated models are used as digital twins to predict the most promising pre-transfer state for each individual starter motor core. We validated the results through additional real-world experiments. Despite some inaccuracies, the results successfully demonstrate the potential of digital twins for predicting probabilistic object transfers in a remanufacturing use-case.

Um die Auswirkungen der globalen Erwärmung zu mildern, haben Regierungen weltweit Ziele zur Reduktion der Treibhausgasemissionen festgelegt. Um diese Ziele zu erreichen, ist eine sozial-ökologische Transformation des Industriesektors von einem "take, make, waste"-Ansatz hin zu einer Kreislaufwirtschaft erforderlich.Das Remanufacturing, eine der so genannten End-of-Life-Strategien, wird häufig als die effektivste Möglichkeit zur Material- und Energieeinsparung angesehen (Nasr and Thurston 2006). Wirtschaftliche Hürden auf Lieferketten und Fabrikebene verhindern jedoch eine breitere Anwendung in verschiedenen Industrien. Eine der größten Herausforderungen beim Remanufacturing ist der Umgang mit der hohen Unsicherheit über den Zustand der zurückgegebenen Produkte. Remanufacturing-Prozesse erfordern eine hohe Prozessflexibilität, die mit den derzeitigen Automatisierungsansätzen nicht erreicht werden kann. Industrie 4.0 Methoden werden häufig als Lösungsansätze zur Reduzierung der Unsicherheiten angeführt. Aus Sicht der Forschung besteht jedoch eine Lücke in der erfolgreichen Demonstration von Implementierungen, die die Vorteile dieser Methoden aufzeigen.

Diese Arbeit trägt dazu bei, diese Forschungslücke zu schließen, indem sie eine reale Anwendung von digitalen Zwillingen im Kontext eines flexiblen Übergabesystems für eine Remanufacturing-Anwendung demonstriert. Der Forschungsbeitrag gliedert sich in vier Teile. Im ersten Schritt wird eine nichthierarchische Architektur für digitale Zwillinge vorgestellt. Darauf folgt die Einführung eines generalisierten Beschreibungsmodells einesÜbergabesystems. Im dritten Schritt wird die reale Implementierung eines Übergabesystems als cyberphysisches System vorgestellt, das sowohl die Architektur als auch das formale Beschreibungsmodell in einer Demonstrationsumgebung umsetzt. In einer umfangreichen Anwendungsstudie werden die digitalen Zwillinge hinsichtlich ihrer Vorhersagefähigkeit für die Übergabe gebrauchter Startermotoren evaluiert.Die vorgestellte Architektur zeichnet sich durch eine nicht-hierarchische Koexistenz von physischen und logischen Assets (einschließlich virtueller Modelle) aus, die durch die Verknüpfung ihrer virtuellen Repräsentationen ein cyberphysisches System bilden. Die Architektur ermöglicht dezentrale Interaktionen zwischen allen Assets und stellt eine gemeinsame Semantik durch die Integration eines Wissensgraphen sicher.

Das verallgemeinerte Beschreibungsmodell eines Übergabesystems basiert im Wesentlichen auf der Beschreibung des Objektzustandes. Zunächst wird der Formierungsprozess des verallgemeinerten Übergabesystems dargestellt. Der Übergabeprozess selbst wird als Funktion von Zustandsübergängen des Objekts abgebildet. Während des Übergabeprozesses wird der Objektzustand kontinuierlich validiert. Dazu werden mehrere kritische Zustandsmengen eingeführt, die zur Steuerung des Übergabesystems verwendet werden. Anhand von Beispielen wird gezeigt, dass das Beschreibungsmodell auf verschiedene Übergabeszenarien angewendet werden kann.

In einer realen Demonstrationsumgebung für die Erforschung von neuen Methoden im Remanufacturing wird ein flexibles Übergabesystem implementiert, das aus einem autonomen mobilen Roboter und mehreren stationären Transfereinheiten besteht. Das System ermöglicht die Übergabe einer Vielzahl von Objekten, ohne dass ein Behälter oder ein anderes Lastaufnahmemittel benötigt wird. Die umgesetzten physischen Assets werden zusätzlich in einer physikalischen Simulationsumgebung modelliert und im cyber-physischen System mit den realen Assets verknüpft.In einer umfangreichen Anwendungsstudie werden sowohl reale als auch virtuelle Experimente für den Transfer von insgesamt 37 verschiedenen gebrauchten Startermotoren durchgeführt. Zunächst werden die virtuellen Modelle anhand einer kleinen Anzahl realer Experimente kalibriert. Dabei ergibt sich ein F0.5-Score von 0,84 für die virtuelle Vorhersage über den Erfolg einer Objektübergabe. Anschließend werden die kalibrierten Modelle als digitale Zwillinge verwendet, um den vielversprechendsten initialen Zustands jedes Startermotors vor der Übergabe vorherzusagen. Die Ergebnisse dieser Vorhersagen werden durch zusätzliche reale Experimente validiert. Trotz einiger Ungenauigkeiten zeigen die Ergebnisse erfolgreich das Potential für den Einsatz digitaler Zwillinge zur Vorhersage unsicherer Objektübergaben in einem Remanufacturing-Szenario.

The integration of both linear and circular processes in one production system poses significant challenges. In particular, the reprocessing of end-of-life products is associated with uncertainties at all levels of the production system, from the initial planning and control through to the executing production hardware and intralogistics. To address these challenges, this article presents approaches for self-learning and autonomously adapting production equipment for the Circular Factory. Initially, hardware and software solutions are developed to cover the necessary processes. Reprocessing is covered by modular and reconfigurable manufacturing cells, which also include new process chains such as the combination of additivesubtractive processes. The provided capabilities must be applied to ever new products, for example by transferring human procedures for unknown products to the production equipment. Lastly, an overall robust and dynamic production planning and control system is developed that maintains continuous operation even in unforeseen situations. The resulting highly dynamic overall system is connected by an autonomous intralogistics system.

In a circular factory, new products are produced reusing parts from used products, as well as newly manufactured parts. The production system consists of disassembly, testing as well as assembly steps. Due to the unforeseeable conditions of the used parts, the complexity of such a circular factory is challenging. This paper contributes a concept of an ontology-based knowledge backbone to master the challenges of such a circular factory. The concept addresses the representation of knowledge especially taking into account uncertainty, how to design queries and means to detect similarities and analogies. Furthermore, the role of research data management with automatized workflows as a supplier for FAIR data is elaborated.

In einer Kreislauffabrik werden neue Produkte unter Wiederverwendung von Teilen gebrauchter Produkte sowie von neu hergestellten Teilen hergestellt. Das Produktionssystem besteht aus Demontage-, Prüf- und Montageschritten. Aufgrund des unvorhersehbaren Zustands der verwendeten Teile ist die Komplexität einer solchen Kreislauffabrik eine Herausforderung. In diesem Beitrag wird ein Konzept für ein ontologiebasiertes Wissensgerüst vorgestellt, um die Herausforderungen einer solchen Kreislauffabrik zu meistern. Das Konzept befasst sich mit der Repräsentation von Wissen unter besonderer Berücksichtigung von Unsicherheit, der Gestaltung von Abfragen und der Erkennung von Ähnlichkeiten und Analogien. Darüber hinaus wird die Rolle des Forschungsdatenmanagements mit automatisierten Workflows als Lieferant für FAIR-Daten herausgearbeitet.

This paper presents a digital twin architecture for a material flow system as well as experiments to determine the sim-to-real gap between virtual models and their real counterparts. The architecture comprises three layers, namely the physical/logical layer, the cyber layer as well as the descriptive layer. Virtual and real assets are able to communicate on the cyber layer by means of OPCUA and the Asset Administration Shell (AAS). The virtual models are implemented in a physics-based simulation environment and allow the reuse of all control and application layer software modules of their real counterpart. In addition, the sim-to-real gap between a selected asset, namely the Transfer Unit is analyzed by means of a collision analysis of the material transfer of simple shaped objects. The results show similar behavior of the real and the virtual models. However, more effort is needed to accurately model the meshes of the collision elements to ensure realistic results.

Individualisation of products and ever-shorter product lifecycles require manufacturing companies to quickly reconfigure their production and adapt to changing requirements. While most of the existing literature focuses on organisational structures or hardware requirements for reconfigurability, requirements and best practices for logical reconfigurations of automated production systems are only sparsely covered. In practice, logical system reconfigurations require adjustments to the software, which is often done manually by experts. With the ongoing automation and digitisation of manufacturing systems in the context of Industry4.0, the need for automated software reconfigurations is increasing. However, heterogeneous and proprietary technologies in the field of industrial automation pose a hurdle to overcome for generally applicable approaches for logical reconfigurations in the industrial domain. Therefore, this paper reviews available technologies that can be used to solve the problem of automated software reconfigurations. For this purpose, an architecture and a procedure are proposed on how to use these technologies for automatic adaptation and virtual commissioning of control software in industrial automation. To demonstrate the interoperability of the approach, collective cloud manufacturing is used as a composing platform. The presented approach further includes a domain-specific capability model for the specification of software artefacts to be generated, allowing jobs to be described and matched on the platform. The core element is a code generator for generating and orchestrating the control code for process execution using the reconfigurable digital twin as a validator on the platform. The approach is evaluated and demonstrated in a real-world use case of a modular disassembly station. © 2023, Publish-Ing in cooperation with TIB - Leibniz Information Centre for Science and Technology University Library. All rights reserved.