Industrial Internet of Things (IIoT) and operational technology (OT) networks are increasingly integrated with physical processes, enabling real-time sensing and actuation while widening the cyber-physical attack surface. Enterprise intrusion detection systems (IDS) often rely on payload inspection, signatures, or high-dimensional engineered features that are difficult to deploy in OT and may yield high false-positive rates while missing some anomalies. This work presents a lightweight IDS for IIoT/OT environments that relies only on passive, per-second traffic telemetry and residual-based forecasting. From mirrored traffic, we extract low-cost statistical features, forming a dataset representation without payload parsing. We emulate denial-ofservice (DoS) flooding and injection/replay manipulations, and use simple rule-based thresholds to label traffic and construct a supervised benchmark dataset. We evaluate two forecastingbased detectors: (i) a supervised Random Forest (RF) forecaster trained on rule-labeled data, and (ii) a one-class Long ShortTerm Memory (LSTM) forecaster trained exclusively on normal traffic (label-free). Real-world experiments on a prototype PLC-to-motor-drive control loop with an external attack platform show that an LSTM-based, residual-driven temporal forecasting approach enables a robust and scalable IDS suitable for timingconstrained IIoT/OT deployments, achieving 100% detection with zero false alarms in the evaluated scenarios.
Multi-robot systems such as cooperative mobile robotic platforms require a high degree of temporal synchronicity to accomplish coordinated tasks with high dynamicity. Due to their mobility, wireless communication technologies such as 5G must be employed. Existing literature presents various approaches for achieving precise wireless time synchronization. However, many of these approaches lack validation in dynamic real-world scenarios that demand a precision of sub-millisecond or even less. To address this gap, this paper presents an experimental demonstration consisting of two independently controlled linear drives, each actuated by as 5G user equipment that are synchronized over the air. For this two different approaches were applied: a fully non invasive state of the art method based on tunneling of Precision Time Protocol (PTP) messages, and a partly invasive synchronization mechanism referred to as the Reference Broadcast Infrastructure Synchronization (RBIS) protocol. The presented demonstration highlights the limitations of PTP as well as the advantages of RBIS-based synchronization for highly dynamic and precise applications.
As a cornerstone of Industrial Cyber-Physical Systems (ICPS) for smart cities, Traffic Accident Early Warning Systems (EWS) are pivotal for maintaining urban resilience. However, reliable accident risk assessment remains elusive due to the extreme “zero-inflated” sparsity of accident data and the complex coupling of heterogeneous urban factors. Conventional methods, often reliant on Recurrent Neural Networks (RNNs) and static graph fusion, struggle with global temporal dependencies and exhibit significant time lags. To address these, this paper proposes AGFTNet (Attention Graph Fusion Spatio-Temporal Network). We introduce a Transformer-based temporal encoder to achieve zero-lag risk perception, overcoming the memory bottlenecks of RNNs. Furthermore, an Adaptive Graph Fusion (AGF) module dynamically learns non-linear interaction weights among Risk, Road Network, and POI graphs, while a risk-sensitive Focal-MSE loss is employed to mitigate bias in sparse data. Extensive experiments on NYC and Chicago datasets demonstrate that AGFTNet significantly outperforms nine state-of-the-art baselines, including GSNet, Graph WaveNet, and STGCN. Specifically, it reduces RMSE by 6.9% and 10.4%, respectively, while maintaining superior recall in high-risk scenarios. The framework offers a robust, low-latency solution for industrial-grade accident risk management.
Industry 4.0 deployments increasingly rely on deterministic communication across heterogeneous Multi-Domain Time Sensitive Networks. However, current management models provide limited observability and cross-domain monitoring. This work-in-progress introduces the Hierarchical Monitoring Orchestrator, an architectural extension to the fully centralized IEEE 802.1Qcc that enables cross-domain telemetry without interfering with the operational state of individual network domains. We propose a standardized YANG model and a twophase signaling procedure to enable real-time metric correlation for guaranteed end-to-end Quality of Service.
Fast and predictable packet processing is essential for time-sensitive PROFINET communication on Linux-based controllers. In this context, the Siemens PROFINET Software Development Kit (PN SDK) provides a widely used toolkit for implementing PROFINET controllers and devices. In current Linux-based deployments, the PN SDK commonly relies on the AF_PACKET path to capture, process and steer Ethernet frames; however, stack-level overhead can limit throughput consistency and robustness under heavy load. Recent Linux kernel technologies based on eBPF and eXpress Data Path (XDP) enable earlier packet handling and efficient packet steering, offering a promising alternative for industrial real-time communication stacks. This paper integrates an XDP-based processing path into the PN SDK controller runtime and provides a comparative evaluation against the AF_PACKET-based variant using (i) iperf3 throughput/jitter measurements and (ii) certificationoriented PROFINET Netload robustness tests. The results show that the XDP-based variant removes the need for stack-level NRT rate limiting in the tested setup and substantially improves timing stability while maintaining PROFINET communication reliability under stress. Across Netload Faulty Communication tests, the XDP variant yields on average approximately 61% fewer AR drops and 49% lower total downtime than AF_PACKET, indicating that early in-driver processing via XDP can improve throughput consistency and robustness for production-grade PROFINET RT controllers.
5G communication system is considered a fundamental part of the Industry 4.0 vision. The escalating energy consumption of 5G Radio Access Networks poses significant challenges. This paper presents a novel approach to improve energy efficiency in 5G Radio Access Networks through a Digital Twin-assisted orchestration framework. Leveraging the Open Radio Access Network, Asset Administration Shell for semantic modeling and Machine Learning techniques, the proposed system enables real-time monitoring, intelligent decision-making, and automated control for dynamic energy saving. Key Performance Indicators and Key Value Indicators are defined and mapped to guide optimization strategies. Experimental evaluation shows the capability of the proposed system to predict energy consumption, consolidate traffic, and manage radio units to reduce power usage without degrading network performance or user Quality of Service.
This paper presents a feasibility study evaluating whether IO-Link Wireless (IOLW) can be implemented in a low-power form suitable for battery-operated industrial sensors while maintaining full real-time capability. Techniques defined in the official IO-Link Wireless specification were analyzed, implemented, and experimentally validated. A complete low-power demonstrator was designed using a Silicon Labs SoC. With optimized communication patterns, reduced message length, selective downlink reception, and efficient sleep scheduling, average power consumption was reduced by up to factor 400, achieving $27.3 \mu \mathrm{W}$ at a mean transmission interval of 60 s. These results confirm the feasibility of IOLW Low Power and reveal required specification updates to ensure consistent interoperability.
Datagram Transport Layer Security (DTLS) is the major security protocol for datagram-based Internet communication, providing retransmission mechanisms to operate over unreliable transport. However, constrained Internet of Things (IoT) links such as Long Range Wide Area Network (LoRaWAN) impose payload limits as low as 51 bytes and duty-cycle holdoffs up to 250 s, exceeding the assumptions of current DTLS IoT profiles and necessitating fragmentation and causing spurious retransmissions or handshake failure. This work-in-progress paper presents DALE (DTLS Adaptive Layer Extension), a framework that enables DTLS in these networks by delegating fragmentation and reliability to the Static Context Header Compression (SCHC) layer and deriving SCHC parameters from link-layer characteristics such as frame payload size and regulatory holdoff. Four experiments comprising ns-3 simulation runs over LoRaWAN Spreading Factors (SF) 7−12 under ideal and fading channel conditions show that DTLS IoT profiles fail at SF9 and above and that retaining flight-level retransmission alongside SCHC fragmentation results in complete handshake failure. Results show that DALE achieves 100% handshake success across all spreading factors on the ideal channel and improves both completion times and success rates under fading conditions compared to existing Internet Engineering Task Force (IETF) timer configurations. An idle-time analysis reveals an SF-dependent tradeoff: at low SFs the DALE timer eliminates spurious retransmissions, while at high SFs it overcompensates with 60-67% idle time, motivating further investigation of adaptive timeout estimation strategies.
Carrying converged IT/OT traffic over a heterogeneous network integrating 5G and TSN links faces two challenges: 5G-introduced jitter, and interference from competing traffic at converged links. This demonstration showcases that scenario with remote PID control of a self-balancing robot over a 5G-TSN network with 6 ms control loop deadlines, which shares a 5G link with video traffic used for an AI-based application. We validate a combined architecture using Hold-and-Forward (H&F) buffers for dejittering and Time-Aware Shaping (TAS) for traffic isolation to enable correct motion control. Five experimental scenarios prove both mechanisms are necessary: disabling either causes robot failure within seconds, while the complete solution enables stable balancing. The robot's visible balance state provides intuitive feedback for validating 5G–TSN integration under millisecond-scale requirements.
Dynamic factory floors require autonomous mobile robot fleets to operate efficiently and reliably in a coordinated fashion. Therefore, this paper presents a decentralized coordination concept that combines local perception with selective, structured information exchange among robots, infrastructure, and supervisory systems. By maintaining a shared situational view, the approach supports adaptive navigation and coordinated task execution while keeping communication overhead low. A live demonstration further illustrates integration with an external dashboard for monitoring and control. The results emphasize the importance of reliable wireless communication as a key enabler for scalable and resilient smart factory operations.
DECT-2020 NR (DECT NR+) represents a paradigm shift in industrial wireless communications as the first non-cellular standard integrated into the IMT-2020 5G framework. Despite its potential for massive IoT and URLLC applications, empirical hardware-level evaluations in representative industrial environments have yet to be conducted. This paper presents a first step towards a physical layer performance characterization of DECT NR+ using pre-commercial evaluation boards interfaced with a real-time RF channel emulator. Following a baseline validation against Additive White Gaussian Noise (AWGN) model, we evaluate system performance across a non-line-of-sight (NLOS) factory profile. While SNR results align with theoretical models, performance in multipath environments suggests that integrating retransmission or diversity mechanisms would further optimize link reliability in dense industrial deployments.
Integrating optical wireless communication (OWC) with powerline enables a cost-effective solution to extend wired industrial networks to the wireless domain. In this paper, we experimentally compare the performance of both uniform and adaptive bitloading for an 1 m optical wireless link with and without powerline communication (PLC) as fixed backbone. In particular, we evaluate the margin-adaptive bitloading algorithm from Fischer and Huber for light-fidelity (LiFi) over PLC. Unlike rate-adaptive algorithms that maximize the throughput, Fischer and Huber algorithm maximizes the signal to noise ratio (SNR) margin (i.e., minimize error probability) for a fixed target rate, making it ideal for industrial applications where reliability is critical. Using a realistic indoor power grid and a line-of-sight (LoS) LiFi link, we analyze the cascaded channel frequency response (CFR), SNR, bit and power allocation and evaluate the resulting bit error rate (BER) and throughput. Experimental results demonstrate that the Fischer and Huber algorithm effectively mitigates the impact of deep spectral notches introduced by the power grid, enabling reliable and robust transmission at data rates of up to 468 Mbit/s.
Reliable wireless connectivity is critical for remote and automated operations in large port terminals. In environments such as the Ponta da Madeira Port (São Luís, Brazil), communication outages in industrial wireless links on mobile stockyard machines may lead to production interruptions and safety risks. Despite the availability of monitoring data, failure detection in industrial wireless networks remains challenging due to the interoperability gap between IT-oriented network monitoring systems and OT-oriented industrial historians. Traditional approaches often rely solely on radio-level indicators and overlook operational context and machine dynamics. This paper addresses these challenges by proposing an interoperability-driven data fusion and harmonization pipeline that integrates wireless monitoring telemetry from Zabbix with operational context from a Plant Information Management System. The pipeline aligns heterogeneous monitoring and operational data into a unified dataset. Wireless failures are labeled from monitoring alarms and modeled as an imbalanced binary classification problem. We evaluate Random Forest, XGBoost and linear Support Vector Machines using imbalance-aware learning on six months of realworld data from 17 stockyard machines. Results show that Random Forest achieves the best performance, with recall above 0.98 while maintaining high precision and balanced F1-scores. Feature importance reveals that operational variables, such as machine position and slewing angle, are stronger predictors of wireless failures than radio metrics. These findings demonstrate that cross-domain data interoperability is essential for robust failure detection in industrial wireless deployments.
Wireless industrial automation systems rely on deterministic communications to support closed-loop control under strict latency and reliability constraints. This paper investigates in-band full-duplex (IBFD) communication as an enabling mechanism for deterministic medium access in industrial wireless networks and analyzes its impact on determinism and control-cycle duration. An IBFD-based superframe is proposed, assuming centralized full-duplex operation at the access point. The proposal is analytically compared with a state-of-the-art deterministic half-duplex baseline, demonstrating reductions in control-cycle time and improved scalability. Numerical evaluations assess the achievable efficiency gains for short-packet, bidirectional traffic characteristic of industrial control loops. Furthermore, the impact of inter-node interference inherent to IBFD operation is also evaluated. The results indicate that IBFD can effectively enhance deterministic communication efficiency in industrial environments while maintaining robustness and scalability.
To address the massive connectivity and low-latency challenges in the Industrial Internet of Things, over-the-air computation has emerged as a promising technique for fast data aggregation. In this paper, a simultaneous multi-function aggregation scheme based on multi-group non-orthogonal multiple access is proposed. This architecture allows for the parallel computation of distinct functions from different sensor groups, effectively supporting multiple active production cells within the same time-frequency resource. Specifically, we formulate a joint optimization problem of power control and scaling factors to minimize the computation mean square error, subject to individual transmit power constraints and minimum data rate requirements for digital metadata transmission. Considering the practical industrial environment, the robustness of the system against imperfect channel state information is explicitly investigated. At the receiver, successive interference cancellation is employed to separate the superimposed signals from different groups. Simulation results demonstrate the effectiveness of the proposed scheme and analyze the trade-off between computation accuracy and the number of supported industrial groups, providing insights for deploying efficient aggregation in dense factory networks.
Bluetooth's Broadcast Isochronous Streams (BISes) enable scalable one-to-many audio dissemination, while Connected Isochronous Streams (CISes) support bidirectional communication between two parties via one-to-one connections. Yet neither is suitable for building ad-hoc communication systems in which a multitude of devices can communicate with each other bidirectionally. BISes are inherently unidirectional, and CISes do not scale. We fill this gap by presenting MANTIS, a mechanism that opens a path toward multiparty communication for Bluetooth LE Audio solely using BISes, despite the fact that this is neither supported nor anticipated by the current Bluetooth Core Specification. By introducing the concept of a token and a coordinated token-passing mechanism, MANTIS transforms inherently unidirectional streams into a flexible system where devices can seamlessly alternate between broadcaster and receiver roles, ensuring a single active broadcaster at any time without interrupting the ongoing, underlying isochronous data stream. Additionally, we implement MANTIS on real-world hardware and validate its operation. By bridging the scalability limitations of CISes and the unidirectionality of BISes, MANTIS introduces a new communication paradigm for Bluetooth, enabling ad-hoc, infrastructure-free audio scenarios previously considered impractical or unattainable.
Machine learning-based intrusion detection for industrial PROFINET networks suffers from a deployment gap: models trained on one site often fail on another because normal timing characteristics vary with factory layout and PLC programs. We present a ground-truth evaluation using 158 packet captures containing 53 confirmed ring-open anomalies from automotive manufacturing facilities across two continents and 53 production lines. We compare eleven detection methods (eight supervised, three unsupervised) using Leave-One-Line-Out cross-validation. ExtraTrees achieves AUROC 0.967 [0.935-0.991] with 88.7% true positive rate at 5% false positive rate. Feature importance analysis shows that all top- 20 features derive from inter-arrival time (IAT) statistics; cycle counters and status fields contribute negligibly. Cross-site transfer experiments reveal false positive rate drift up to 30% (6× the 5% target), demonstrating that domain-aware threshold calibration is essential for deployment.
The Industry 4.0 strategy was born in Germany almost 15 years ago. It envisioned industrial systems that are autonomous, flexible, and capable of seamless collaboration across heterogeneous components. Despite significant technological advances, the original vision remains largely unrealized. A key limiting factor is the absence of interoperability across systems and their components. In this paper we argue that while physical connectivity has improved, alignment of semantics has not kept pace, leading to fragmented architectures, incompatible data formats, and costly custom integrations. We illustrate these problems with a use case, focused on the lack of interoperability of clock synchronization in systems-of-systems. We propose a new paradigm for interoperability, elevating it to the status of a life-cycle property, on par with safety, reliability, and realtime. We discuss the need for continuous monitoring, governance, and adaptation. By reconceptualizing interoperability as a longterm, dynamic system attribute, we outline a structured life-cycle model that supports sustained semantic consistency and enables resilient, future-proof industrial automation ecosystems.
The integration of Fifth Generation Mobile Network (5G) and Time Sensitive Networking (TSN) is a key enabler for future industrial communication networks, combining deterministic Ethernet with dynamic wireless connectivity for use cases such as modular manufacturing. While standardization efforts lay the foundation, the actual realization of 5G as a logical TSN bridge requires consistent configuration models and Quality of Service (QoS) enforcement across both the wireless and the wired domains. In particular, challenges arise from the collection and generation of metrics that enable configuration, the transport of these metrics through the entire system, and the mapping of TSN parameters to 5G-specific resource allocation mechanisms. In addition, the architectural characterisitics and constraints within current tool chains provided for these network management purposes need to be considered. From a research standpoint, such integration is a challenge for the scalability and orchestration of hybrid networks in real-world deployments. Building on this analysis, the paper proposes a configuration concept leveraging standardized data models and existing solution modules. For the validation of the concept a dual testbed is presented which combines Centralized Network Controller (CNC)-based TSN configuration with Open5GS/ UERANSIM-based 5G simulation.
Secure end-to-end communication is essential for IoT deployments, yet Transport Layer Security (TLS) proves unsuitable for Low Power Wide Area Network (LPWAN) networks due to high overhead and connection-oriented design. Datagram Transport Layer Security (DTLS) addresses these limitations through datagram-based secure communication over UDP, but constrained networks require additional optimizations to achieve practical performance. This paper presents an implementation and evaluation of DTLS version 1.2 with Static Context Header Compression (SCHC) over commercial LoRaWAN infrastructure using severely constrained hardware with 256 kB flash memory and 64 kB random-access memory. The implementation employs context-based compression for IPv6, UDP, and DTLS headers. Controlled testbed experiments quantify handshake duration, compression efficiency, energy consumption, and resource utilization across spreading factors, acknowledgment modes, and authentication methods. Results demonstrate that header compression reduces handshake duration by 41.7%. Certificatebased authentication requires 146.9 s at spreading factor 12 compared to 73 s for pre-shared keys, while requiring 95.7% of available flash memory. Spreading factor choice has the largest performance impact, causing a 65.2% variation between SF7 and SF12. The work quantifies performance limits for secure LPWAN applications, showing that real hardware introduces extra handshake delays beyond those predicted by simulation.
