Battery production consists of tightly interconnected process steps, yet physical inspections are performed only at selected points due to cost and time constraints, causing late scrap detection and unnecessary resource use. With increasing data availability, virtual measurements can predict quality from process data without additional inspection effort. However, existing approaches lack uncertainty quantification comparable with measurement uncertainty and compatible with established metrological standards, while remaining flexible in algorithm choice. This paper proposes a framework for virtual measurements in multistage production, enabling quality prediction at each process step and quality propagation between consecutive process steps. Uncertainty is quantified using ensemble methods, interpreted as an empirical measurement model combined with a Monte Carlo approach, enabling uncertainty estimates comparable to the GUM and allowing existing standards for conformity assessment and test process capability to be applied. By providing uncertainty-aware virtual measurements at every production step, the framework enables early, standards-compliant quality decisions without additional physical inspections. The framework is validated on 1,000 synthetic battery modules across three production steps. The best virtual measurement achieves R-2 = 0.87 at the stacking stage, with a total uncertainty of 0.030 W versus a reference of 0.024 W, demonstrating uncertainty-aware forecasting and avoidance of overconfident decisions.

The aerospace industry’s drive to improve fuel efficiency and reduce emissions has led to the widespread adoption of lightweight, thin-walled structures in aircraft assemblies. However, these flexible structures are susceptible to deformation under gravitational and external forces, causing assembly deviations and potential misalignment. To counteract the misalignment caused by part deformation in joining processes, shims are specifically designed to fill the gaps between components. Yet, despite advances in metrology, the process of shimming remains largely manual, since each assembly’s unique gap profile requires physical measurement, fitting, and adjustment, which hinders full automation. A promising solution is the use of a digital twin to simulate and predict part deformation during the joining process. The predicted deformation resulting from the digitalization of the compliant structures to be joined can be used for designing shims in a virtual assembly environment or as input for adaptive tooling systems specifically designed to compensate for the deformation. This work introduces a concept for a test cell equipped with a large-scale metrology system to capture the data needed for constructing a digital twin of manufactured large, compliant, thin-walled structures. Corresponding data pipelines, designed to meet FAIR data principles, combine a time-series database for continuous measurement data with a non-relational database for storing measurement metadata. Preliminary results demonstrate the test cell’s ability to provide the data required for applying inverse form finding methods to approximate the LCTS gravity-free state shape and for training data-driven physics informed models for the prediction of the deformation behavior. These findings lay the groundwork for future digital twin development and offer insights to improve alignment precision in the assembly of large, compliant, thin-walled structures.

The finite nature of the Earth’s resources and the need to reduce CO2 emissions require a shift from a linear to a circular production. Remanufacturing is a circular economy strategy that involves inspecting, disassembling, and reassembling of returned products to ensure they have equal or higher functionality before being dispatched again. However, the quantity and quality of returned products are uncertain, meaning production systems must be flexible enough to adapt their assembly and disassembly processes accordingly. Furthermore, products are designed for efficient assembly, but not for the flexible assembly and disassembly required. Currently, planning assembly and disassembly sequences is a manual and tedious task, and the resulting sequences are inflexible. We reviewed current approaches to derive assembly sequences automatically from CAD files. Based on this, we developed an approach that derives an AND-OR graph from STEP files. We then fed the resulting flexible assembly sequences into a discrete event simulation of flexible, hybrid assembly systems. The simulation results show diverse production KPIs based on different scenarios, which product designers can use to compare the impact of different design decisions on flexible, hybrid disassembly and assembly. In the future, we plan to deploy reinforcement learning to further automate design decisions.

The increasing product variability and skilled labor shortages in manufacturing intensify the need for more flexible and adaptive automation solutions, particularly in assembly systems. However, existing robotic automation typically requires expert-level programming, preventing adaptability to new tasks or product variants. This work presents a modular embodied AI agent framework that enables non-programming operators to implement and reconfigure robotic tasks through intuitive natural language (NL) commands in text or voice format. By integrating open-source Large Language Models (LLMs) and Visual Language Models (VLMs) with a Robot Operating System 2 (ROS2)-based stack, the agent translates user instructions into perception, grasp generation, and motion planning actions. The framework is evaluated in real-world experiments across four scenarios: low-level motion control, vision-guided cable grasping, visual scene feedback, and task-cycle recording with replay for scalable execution. Results show 100% command parsing accuracy for basic motions, reliable visual feedback, and 70% success in dual-cable manipulation, with failures mainly from trajectory planning or self-collisions. VLM inference latency proved highly hardware-dependent, with near real-time performance on high-end GPUs but significant slowdowns on consumer devices, highlighting deployment challenges. These findings show the potential of embodied AI agents to bridge NL interaction with robotic execution, lowering barriers to deploying adaptive, operator-friendly robotic systems in dynamic manufacturing.

The wide-spread adoption of large thin-walled structures made of carbon fiber reinforced polymers (CFRP) in the aeronautical sector contributes to better fuel efficiency and lower CO2 emissions. However, despite their advantages, the automation of assembly processes involving these structures remains challenging due to their compliance which translates into a tendency to deform under process induced loads. Model-based control strategies can mitigate deformations during assembly, but high-fidelity simulations such as finite element models (FEM), though accurate, are too slow for real-time integration. While accurate, these simulations can be computationally expensive and time-consuming, particularly for large-scale problems or when multiple design iterations are required. This paper presents surrogate modeling approaches to approximate the mechanical response of large compliant thin-walled CFRP structures. The proposed work presents a benchmark study of different surrogate modeling techniques, ranging from classical machine learning methods to physics-informed neural networks utilizing the DeepFEM framework. The performance of these surrogate models is evaluated based on their accuracy, computational efficiency and ability to generalize to unseen data. The results demonstrate the potential of surrogate modeling as a viable alternative to high-fidelity simulations based on FEM, enabling the integration of model-based control strategies in the assembly of large CFRP structures. Key findings show that kernel-based methods like RBF and GP are accurate but computationally prohibitive at larger scales, whereas MLP models compute 200 test cases in under 0.1 seconds and scale efficiently, making them suitable for real-time or integrated process applications.

Adaptive robots in dynamic production environments require robust perception capabilities, including 6D pose estimation and multi-object tracking. To address limitations in real-world data dependency, noise robustness, and spatiotemporal consistency, a LiDAR framework based on the Robot Operating System integrating a synthetic-data-trained Transformation-Equivariant 3D Detection with multi-object-tracking leveraging center poses is proposed. Validated across 72 scenarios with motion capture technology, overall results yield an Intersection over Union of 62.6

The increasing diversity and shorter life cycles of technical products pose significant challenges for manufacturing companies, particularly in the context of providing specific and context-sensitive instructions to employees, especially in domains including maintenance, assembly and disassembly. This challenge holds significant importance in the context of the current skilled worker shortage. This paper proposes a solution by leveraging digital twin technology and smart services to automate the generation of context-sensitive instructions. The research outlines the development of a smart service system that uses real-time data from digital twins to create and deliver adaptive and user-specific instructions via smart devices. A conceptual design of the smart service system, a prototypical implementation using a rolling mill maintenance task, and the verification and validation of the developed system were carried out. The results indicate that the proposed system effectively addresses the challenges of traditional manual instructions, enhancing efficiency, accuracy, and user satisfaction.

With the growing reliance on deep learning (DL) models for tasks in computer vision, high-quality training data is essential for the development of accurate models. However, generating large, labeled datasets from real-world sources is time-consuming, laborious, and expensive. Synthetic datasets offer an alternative, allowing for the automated generation of large amounts of training data. Despite their advantages, synthetic datasets often fail to capture the complexity of real-world images, leading to a performance gap when DL models trained on synthetic data are applied to real-world tasks. To improve the performance of synthetically trained DL models, it is crucial to identify factors that contribute to the domain gap between synthetic and real-world datasets first. Yet, the high dimensionality of image data yields a challenge that impedes a direct comparison between image datasets. This work explores the utilization of the UMAP algorithm to assess the similarity between synthetic and real-world image datasets through their embedding and analysis in a low-dimensional space.

Although digital twins are increasingly used for pre-deployment testing, their reliability as predictive tools remains understudied due to the lack of established validation frameworks. This paper presents a systematic methodology for validating the predictive fidelity of physics-based digital twins in robotic navigation tasks, addressing a critical gap in sim-to-real transferability for industrial mobile manipulators. We propose a novel evaluation approach combining (1) multi-metric comparison (localization accuracy, path consistency, goal accuracy, and navigation performance) between real-world and simulated navigation experiments, and (2) an uncertainty quantification method to establish confidence intervals for digital twin predictions. Using an NVIDIA Isaac Sim model of an omnidirectional mobile manipulator and digitally reconstructed production environments, we conduct 50 real-world and 50 digital twin experiments across five industrial scenarios. The results show a mean Hausdorff distance of 0.195 m between real and simulated paths, localization RMSE differences of 0.005 m, and a path prediction accuracy of ±0.229 m (95

Effective decision-making in automation equipment selection is critical for reducing ramp-up time and maintaining production quality, especially in the face of increasing product variation and market demands. However, limited expertise and resource constraints often result in inefficiencies during the ramp-up phase when new products are integrated into production lines. Existing methods often lack structured and tailored solutions to support automation engineers in reducing ramp-up time, leading to compromises in quality. This research investigates whether large-language models (LLMs), combined with Retrieval-Augmented Generation (RAG), can assist in streamlining equipment selection in ramp-up planning. We propose a factual-driven copilot integrating LLMs with structured and semi-structured knowledge retrieval for three component types (robots, feeders and vision systems), providing a guided and traceable state-machine process for decision-making in automation equipment selection. The system was demonstrated to an industrial partner, who tested it on three internal use-cases. Their feedback affirmed its capability to provide logical and actionable recommendations for automation equipment. More specifically, among 22 equipment prompts analyzed, 19 involved selecting the correct equipment while considering most requirements, and in 6 cases, all requirements were fully met.

The current research presents an extension of the Pit-Stop Manufacturing framework. It addresses the challenges of managing complexity and uncertainty in the production ramp-up phase of manufacturing systems, bridging the gap in existing approaches that lack comprehensive, quantitative, and system-level solutions. This research integrates state-of-the-art methodologies, utilising such metrics as Overall Equipment Effectiveness and Effective Throughput Loss to enhance ramp-up management. The developed framework is represented by a conceptual model, which is translated into a digital product combining multiple artefacts for comprehensive ramp-up research, namely a digital twin of the production system, a Custom Experiment Manager for multiple simulation runs, and a Graph Solver that uses the stochastic dynamic programming approach to address the decision-making issues during the production system ramp-up evolution. This work provides a robust decision-support tool to optimise production transitions under dynamic conditions by combining stochastic dynamic programming and discrete event simulation. The framework enables manufacturers to model, simulate, and optimise system evolution, reducing throughput losses, improving equipment efficiency, and enhancing decision-making precision. This paper demonstrates the framework’s potential to streamline ramp-up processes and boost competitiveness in volatile manufacturing environments.

Die Transformation von einer linearen zu einer zirkulären Produktionswirtschaft ist notwendig, um CO 2 -Emissionen zu reduzieren und Primärressourcen zu schonen. Flexible Montagesysteme können eine effiziente De- und Remontage ermöglichen, erfordern jedoch innovative Planungsmethoden. Ein neues Vorgehensmodell bewertet die Demontierbarkeit von Produkten bereits in der Entwicklungsphase mithilfe von CAD-Daten und Simulation. Dies unterstützt technische Produktdesigner bei der Gestaltung kreislauffähiger Produkte.

Manufacturing companies need to produce increasingly individualised products in shorter time to stay competitive. However, production ramp-ups are often poorly executed and fail to meet their targets. During ramp-up, there is a lack of sufficient data to base decisions on and managers fly blind. The emergence of Industry 4.0 promised real-time data-based decision-making but is not yet exploited during production ramp-up due to this data gap. This systematic literature review addresses this gap by identifying methods since 2011 to close this production ramp-up data gap by following the Preferred Reporting Items for Systematic reviews and Meta-Analyses statement. The reports included in the review are categorised according to manufacturing process, ramp-up field of action, and production system level. We found a lack of designated data models for production ramp-up, mostly unsystematic approaches for dealing with small amounts of data and general barriers to deploying Industry 4.0 solutions during the ramp-up phase. Finally, we infer that the research and industry communities should share more data and models to validate systems and tools.

Circular economies require efficient disassembly processes for end-of-life products from various product series, variants and with various product states. CAD-based Disassembly Sequence Planning aims to efficiently create feasible disassembly sequences to minimize the needed effort in disassembly execution. Manual, collision-based and feature-based Disassembly Sequence Planning approaches are predominant in research. However, there is no tool covering multiple of these techniques simultaneously. In this paper, we aim to enable the combination of these techniques in a parallelized and hybrid manner. A Service-Oriented Model-View-Controller Framework is presented as an overarching software architecture. The architecture was implemented and tested on an exemplary use case. The validation proved the potential quality and efficiency increase of CAD-based disassembly sequence planning. The framework contributes to choosing the right disassembly sequence planning strategy and thus increasing the efficiency of the circular economy.

Abstract In diesem Beitrag wird ein neuartiges Konzept für ein hybrides Produktionssystem vorgestellt, das sowohl neue als auch aufbereitete Produkte verarbeiten kann. Im Kern zeichnet es sich durch einen multifunktionalen De- und Remontagearbeitsplatz aus, der mehrere Strategien der Kreislaufwirtschaft in das Wertschöpfungssystem integriert. Damit wird das Ziel verfolgt, die industrielle Transformation hin zu einer kreislauffähigen Produktion zu fördern und effizient zu gestalten.

Manufacturing systems are undergoing systematic change facing the trade-off between the customer's needs and the economic and ecological pressure. Especially assembly systems must be more flexible due to many product generations or unpredictable material and demand fluctuations. As a solution line-less mobile assembly systems implement flexible job routes through movable multi-purpose resources and flexible transportation systems. Moreover, a completely reactive rearrangeable layout with mobile resources enables reconfigurations without interrupting production. A scheduling that can handle the complexity of dynamic events is necessary to plan job routes and control transportation in such an assembly system. Conventional approaches for this control task require exponentially rising computational capacities with increasing problem sizes. Therefore, the contribution of this work is an algorithm to dynamically solve the integrated problem of layout optimization and scheduling in line-less mobile assembly systems. The proposed multi agent deep reinforcement learning algorithm uses proximal policy optimization and consists of a decoder and encoder, allowing for various-sized system state descriptions. A simulation study shows that the proposed algorithm performs better in 78% of the scenarios compared to a random agent regarding the makespan optimization objective. This allows for adaptive optimization of line-less mobile assembly systems that can face global challenges.

In light of the challenges posed by the often unavailability of coherent data in manufacturing for operational Artificial Intelligence (AI) decision support systems, the generation and utilization of synthetic datasets have become essential. This study introduces a simple numerical Synthetic Simulated Environment (SSE) using timed and parametrizable Petri Net (PN) modules, embedded in a Directed Acyclic Graph (DAG) structure described by an adjacency matrix to represent material flow. Implemented in PyTorch for seamless integration with AI components, our simulation framework simplifies manufacturing systems, yet remains expandable for diverse use cases. The simulation model was demonstrated displaying its capability of generating synthetic data. This approach explores the practicality and applicability of generated data. It could serve as an ideal environment to benchmark Artificial Intelligence (AI) algorithms in comparative experiments, investigating operational problems featured in the dynamic interactions of discrete manufacturing systems.

In response to the shortage of skilled workers and the increasing diversity of products and variants, integrating adaptive and flexible mobile manipulators into line-less mobile assembly systems (LMAS) stands out as a promising solution. To fully benefit from these robotic systems in LMAS, offloading computationally intensive tasks from mobile manipulators to edge computing systems is essential. This offloading of computational tasks not only lightens the processing load on individual robots but also enhances AI based perception, improving the overall system capabilities in dynamic manufacturing environments. Achieving these goals requires a robotic edge computing framework that serves as the key to flexible and adaptive mobile manipulation in LMAS. To construct such an edge computing framework, a comprehensive set of building blocks is required. These building blocks include a powerful edge system with high processing power, sensor-equipped mobile manipulators with the ability to perceive the environment, and a robust edge-to-robot connectivity. A comprehensive software stack, including software modules for localization and navigation in dynamic assembly environments, AI perception for precise pose estimation of assembly components, and motion planning for interaction between manipulators and assembly components, is crucial. Furthermore, virtualization techniques, a comprehensive deployment strategy, and a detailed description of robot hardware, site, and resources are essential. The proposed edge computing framework presents a solution that addresses mobile manipulators in LMAS and paves the way towards advanced industrial automation. Through implementation in a research assembly environment, the realized proof-of-concept showcased the feasibility of the proposed edge computing framework in a real-world pick-and-place scenario, highlighting its potential to enhance adaptivity and flexibility. There is a potential to refine and expand the framework to accommodate a broader range of industrial applications and streamline diverse manufacturing processes through the integration of mobile manipulators and edge computing within industrial settings. (c) 2024 The Authors. Published by Elsevier B.V.

Global trends such as mass customization and dynamical disturbances in manufacturing systems demand robustness for steady and efficient production control. These dynamical disturbances significantly affect scheduling as a central task of production control. The task allocation procedure is designed reactively in online scheduling, making it suitable for application in dynamic manufacturing environments. Recently, the usability and accuracy of machine learning (ML) methods improved significantly, leveraging online scheduling performance to deal with the above-described upcoming challenges in manufacturing. Since no up-to-date reviews exist that focus on machine learning in online scheduling, this paper presents a systematic and specific literature review. The review was conducted according to the PRISMA meta-analysis framework to ensure repeatability. The findings reveal the diverse applicability and performance of supervised, unsupervised, and reinforcement learning techniques in various manufacturing scenarios, aiding researchers and practitioners in the selection and deployment of ML methods. Moreover, the review identifies future research trends in ML methods for online scheduling applications, aligning with the overarching goal of accelerating manufacturing processes.