Mixed zonal and domain-oriented automotive E/E architectures distribute cause-effect chains across fieldbus segments and automotive Ethernet (ETH). To bridge this multi-network communication, a tunnel mechanism bundles multiple fieldbus PDUs into shared ETH frames and routes them via a central PDU router, reducing overhead and processing load. However, bundling introduces packing delay, jitter, and flow discretization, and generates bursts when PDUs are unpacked onto lower-rate fieldbuses. These effects create cross-interactions between cause-effect chains and delay unrelated traffic. This paper identifies and describes the problem from a system-wide perspective and outlines ongoing work on optimization strategies.
Virtual commissioning (VC) of special purpose assembly and handling machinery is a mature technique but remains hard to scale, as no clear model structure links the heterogeneous engineering data across disciplines into a coherent simulation model, which hinders automated VC model generation. This paper proposes such a structure based on VDI 2860 handling functions, which form a semantic link across four engineering views covering geometry, electrics, software, and simulation and thereby remain flexible enough for strong variant diversity. The structure is populated automatically from computer aided design (CAD) data by identifying simulation relevant bodies through geometric analysis and classifying them deterministically by their functional effect rather than by their geometric shape, while data from the electrical and software views complement the parameterization in subsequent steps. Evaluation on a real special purpose assembly confirms robust application programming interface (API) based data reconstruction and reliable function based classification, providing an automatable path from CAD data to parameterized and runnable simulation primitives for automatic VC model generation.
The increasing complexity of Electrical Substations, driven by the integration of renewable energy sources and the demand for real-time monitoring and control, has reinforced the role of Substation Automation Systems (SASs) compliant with the IEC 61850 standard. In this context, communication requirements within the Substation Communication Network have become more stringent, promoting the adoption of Time-Sensitive Networking (TSN) due to its deterministic capabilities, time synchronisation, and efficient resource management. In line with these challenges, this work proposes a novel platform for testing TSN usage at the station level of the SAS. The design is compliant with IEC 61850 requirements, particularly regarding space redundancy and clock synchronisation, and incorporates realistic traffic patterns and a topology inspired by High-availability Seamless Redundancy (HSR). Additionally, representative traffic flows are defined, and the integration of the Frame Replication and Elimination for Reliability (FRER) mechanism and clock synchronisation are analysed.
The advent of Industry 5.0 emphasizes human-centricity in autonomous systems, making safe Human-Robot Collaboration (HRC) a critical challenge. Most of the solutions to guarantee safe HRC rely on equipping robots with custom sensors and control systems, resulting in poor portability across different robotic platforms and a narrow focus on the robot's immediate surroundings, lacking broader environmental awareness. This paper describes the design and safety engineering of CogniSafe3D, a novel system that ensures safe HRC by leveraging a static 3D LiDAR sensor monitoring the collaborative environment. Unlike state-of-the-art methods, CogniSafe3D requires no sensing hardware on the robot nor on human collaborators, interfacing with the robot solely via an API to trigger safety actions and dynamic path planning. This paper reports our experience in the design of the CogniSafe3D model, following a standard-compliant safety process. While applying the process, the need to detect specific corruption in the LiDAR's point cloud data emerged as mitigation of high-risk hazards, leading to the development and evaluation of a point cloud anomaly detector, which is discussed and whose experimental details are publicly available for reproducibility.
Reliable robotic manipulation of tiny industrial parts requires accurate geometric perception. However, acquiring high-quality 3D data at the millimeter scale cost-effectively remains challenging and is largely underrepresented in existing 3D reconstruction datasets. In this work-in-progress paper, we introduce TIP3D, a dataset designed to benchmark passive image-based 3D reconstruction methods on real-world tiny industrial components. The dataset currently contains 360° video recordings of 26 different tiny industrial parts captured in multiple configurations. For each scene, high-precision laser scans provide geometric ground truth. In addition, we propose a classification scheme that categorizes parts according to factors known to influence 3D reconstruction performance. We further outline a evaluation pipeline that aligns reconstructed point clouds with the ground truth and evaluates geometric accuracy using established metrics. Our experiments with modern reconstruction methods demonstrate that millimeter-scale industrial components present significant challenges for current approaches.
