This paper addresses the need for an indoor-focused, easy-to-replicate road-sign dataset that captures unprocessed raw image data. Existing datasets typically focus on processed RGB images, which limits their usefulness for research on embedded, end-to-end machine learning. To fill this gap, a Raspberry Pi Camera Module v1.3 was mounted on autonomous mini robots, which ran continuously in various indoor settings. Over a period of two months, approximately 70,000 10-bit, 5-megapixel images were stored as TIFF files. Exposure and ISO were intentionally varied to introduce motion blur, noise, overexposure, and underexposure. The resulting labeled dataset comprises around 47,000 bounding boxes for 87 sign categories, including danger, regulatory, directional, demo-specific, and unknown signs. This work provides a compact, low-cost framework that enables researchers and educators to explore algorithms on raw images in a reproducible indoor setting through both long-term data collection and in-classroom demonstrations.

The integration of data-driven models and specifically machine learning for conditon monitoring and predictive maintenance into companies, especially small and medium-sized enterprises, offers significant opportunities in reducing costs, operating more sustainably, and maintaining long-term competitiveness. However, many small and medium-sized enterprises lack the necessary resources and expertise to derive knowledge from data and integrate their own machine learning based solutions. To address this challenge, a framework is presented that enables the automated generation of data-driven models with a particular focus on condition monitoring and predictive maintenance, but applicable to other use cases as well. Using a dataset from the 2022 data challenge of the prognostics and health management society, it is demonstrated that the framework can generate high-performing models, achieving F1-scores up to 0.998, exemplarily for a classification task.

Although digital twins are increasingly being used to represent physical assets in industrial automation, most of them remain passive, merely building a digital shadow of the asset. Their potential as active, autonomous components in cognitive control architectures remains largely unexplored. This paper presents a novel approach to realize executable digital twins within cognitive operators by embedding the Sense–Plan–Act paradigm into standardized submodels of the Asset Administration Shell. Specifically, the submodel Time Series Data is used to capture dynamic system state for the sensing phase, while the Asset Interfaces Description represents executable interactions for the planning phase. By enabling each assets’ cognitive operator to interpret these submodels, distributed systems can reason and act through their digital representation, while maintaining semantic interoperability and compliance with standards. The concept is validated in a decentralized task allocation scenario using modular autonomous robots. Initial results confirm the technical feasibility and reusability of the approach and highlight the potential of semantically enriched digital twins in future industrial systems.

Klarhorst C, Quirin D, Hesse M, Rückert U. A Digital Twin Implementation for the AMiRo. In: 2024 IEEE 29th International Conference on Emerging Technologies and Factory Automation (ETFA). IEEE; 2024: 1-4.

The holistic application of data-driven value-adding services (DDSs) to shop-floor assets (SFAs; e.g. equipment, machines, components) is a major obstacle, particularly for small and medium-sized enterprises (SMEs): Due to the high complexity and required expertise in various disciplines. It is necessary to i) select suitable DDSs for a relevant component in production, ii) provide data for their training, iii) deploy the DDSs in the given IT/OT infrastructure, and iv) setup data-streams between SFAs, DDSs, and dashboards. Currently, many companies are implementing flagship projects that go through the necessary steps for individual components, but these isolated solutions are not scalable due to the need for manual intervention. This paper describes how to completely automatize the four steps and their consolidation in a scalable Industry 4.0 one-stop-shop solution. To achieve this goal, each asset is equipped with an Asset Administration Shell (AAS) that specifies, in particular, the represented asset's capabilities, the data required to apply such a capability and the data generated by a capability. Based on the AASs of SFAs, DDSs, and computing resources (CRs), a matching service suggests automatically which DDSs would add value to which SFAs and, for each SFA-DDS match, in which CR to deploy the DDS. A trade fair demonstration provides strong support for the statements made: The one-stop-shop solution including the automatic matching greatly simplifies the application of data-driven services and, in some cases, even makes it possible at all.

One necessary part of Industry 4.0 is the availability and accessibility of data processing pipelines. This paper shows the ongoing development of ML4ProFlow, a framework that brings together the following parts: First, it provides the management of execution environments. Second, it specifies processing modules that focus on reusability and cross-platform usage. Third, it comes with a benchmarking automation to help developers implementing and analyzing modules and their combination. Those three integral parts of the framework are presented and the usability is shown.

Today's high complexity and required expertise in various disciplines for data-based evaluations of shop-floor assets is challenging. This paper describes the ongoing development towards an Industry 4.0 ecosystem enabling Smart Services and shop-floor assets to network autonomously. Three partial solutions are combined for this purpose: Industry 4.0 digital twins, automated data streams and a Smart Service toolbox. A prototypical implementation proves the general practicability. Furthermore, future work is outlined to achieve full autonomy.