The evaluation of Cyber-Physical Production Systems (CPPS) using Digital Twin (DT) architectures and Asset Administration Shells (AAS) is critical for advancing flexible, autonomous, and human-centric manufacturing environments. However, small and medium-sized enterprises (SMEs), particularly in Latin America, face significant barriers to adopting and evaluating these technologies due to resource limitations, complexity, and lack of standardized frameworks. This paper proposes a scalable and modular evaluation framework specifically designed for SMEs, integrating DTs for real-time simulation and monitoring, AAS for semantic interoperability, and simulation-based assessment methods to measure CPPS autonomy, flexibility, and resilience. A comprehensive review of recent approaches, trends, challenges, and research gaps highlights the urgent need for dynamic, standardized, and cognitive-enabled evaluation methodologies. The proposed framework addresses these needs by formalizing production knowledge through ontologies, synchronizing physical systems with digital models, and enabling systematic performance evaluation across diverse production scenarios. A case study conducted in a Latin American manufacturing SME demonstrates the practical application of the framework, showing improvements in production efficiency, decision transparency, and system adaptability. The results underline the framework’s potential to democratize access to advanced Industry 4.0 solutions for SMEs, fostering more resilient and intelligent manufacturing ecosystems. By bridging current technological gaps, this work contributes to accelerating the digital transformation of SMEs and supports the broader goals of Industry 5.0. Future research will explore the integration of explainable AI techniques and expand the framework’s applicability to multi-site production networks.
The Performance-driven PID (Pd-PID) control method, which utilizes the minimum variance control index (MV-Index), has garnered significant attention. In Pd-PID, a deadbeat control system is always considered optimal for evaluation, regardless of the system’s response speed. As a result, systems with slower responses receive lower evaluations, even when good control performance can be obtained. Furthermore, the Pd-PID method continuously updates PID parameters to improve control performance evaluation, which can ultimately lead to system instability. This paper addresses this issue and proposes a new control performance evaluation index to replace the MV-Index.
Large Language Models (LLMs) have shown promising capabilities in supporting and automating a variety of clearly defined, structured tasks. The engineering of modular plants also follows standardized workflow structures and uses standardized artifacts such as P&IDs, flow diagrams or HAZOPs. Therefore, the engineering of modular plants has the potential to be significantly supported or automated by AI tools. The articles objective is to determine which aspects of the modular plant engineering workflow could be supported or automated by LLMs. This is done by reviewing the current state of the art and combining the modular plant engineering workflow which LLM-based capabilities for engineering support. This leads to two results. First, a mapping of current LLM capabilities to the steps within the modular plant engineering workflow. Second, a theoretical assessment of the general LLM capabilities in terms of their usefulness for the tasks within the modular plant engineering workflow. All in all, the proposed mapping of LLMs to the modular plant engineering workflow provides an overview of the ways in which LLMs can support engineers. In addition, the LLM capability assessment provides insights into the current state of the LLMs abilities and highlights where further research and development is needed.
The structural assessment of submerged cathodic protection systems in Offshore Wind Turbines (OWTs) is crucial for ensuring longevity and operational efficiency. Traditional underwater inspections are expensive, inefficient, and expose human divers to hazardous conditions.This article aims to enhance the perception capabilities of underwater vehicles by introducing the Contextual Anode Locator in Varying Underwater Scenarios (CALVUS), a learning-based solution designed for the robust and precise detection of sacrificial anodes in harsh subsea environments. CALVUS leverages the feature extraction capabilities of a depth estimation ViT-based backbone to detect anode structures under challenging underwater conditions such as heavy marine snow, variable illumination, biofouling and motion blur.Evaluation on a dataset composed of images captured at the ATLANTIS Test Centre, CALVUS shows a performance of AP@50 of 97.9 %, an improvement of 19.9 % over state-of-the-art networks such as YOLO and RT-DETR. These results demonstrate the added value of using depth features during the detection operation, ultimately contributing to improved OWT operational efficiency and reduced maintenance costs.
Ensuring high-quality data is essential for effective air quality monitoring, especially in smart urban environments, where real-time insights drive health and policy decisions. This paper presents a structured, step-by-step approach for developing FIWARE-based air quality monitoring systems with a strong focus on data quality. Addressing challenges, such as sensor heterogeneity, calibration, real-time data processing, and semantic interoperability, the paper outlines best practices for achieving accurate and reliable environmental data. The system design leverages open-source FIWARE components integrated with NGSI-LD Smart Data Models to ensure standardized data flows. Furthermore, the TwinAIR project is explored as a real-world implementation that enhances data quality through validation and digital twins for air quality trend prediction. The proposed architecture demonstrates how high data integrity can be maintained across the entire data life-cycle — from sensor deployment to actionable analytics. This work is a practical guide for system designers and a benchmark to achieve scalable, interoperable, and reliable air quality monitoring solutions.
IoT networks require efficient intrusion detection systems that balance detection accuracy with computational constraints for edge deployment. This work introduces a novel packet-level intrusion detection method using 14 statistical features extracted from sliding windows through basic counting operations. The approach operates directly on fixed-size packet windows without requiring flow reconstruction or deep packet inspection to compute statistics. We compare this packet-level method against traditional flow-based feature extraction using tstat on the Edge-IIoTset dataset across 10 attack types. Both approaches train One-Class Support Vector Machine (OCSVM) models for fair evaluation. Results demonstrate that our packet-level method achieves superior detection rates with lower false alarms while significantly reducing feature extraction time compared to traditional flow-based methods.
Resource sharing in multicore architectures remains a significant challenge that limits the performance potential of such systems, especially in industrial domains requiring high performance and strict timing guarantees. In this paper, we introduce a novel synchronization protocol called Multi-Replicas Time-Bounded Consistency (MTC), designed to mitigate the adverse effects of resource sharing among tasks distributed across different cores. MTC achieves this by replicating shared resources while maintaining a bounded level of consistency among replicas within a defined time constraint. We present a response-time analysis of the proposed protocol and demonstrate that MTC can substantially simplify the evaluation of response-time analysis compared to traditional lock-based synchronization methods, offering a potentially more efficient and scalable alternative.
In this paper, we propose a computationally efficient quadratic programming (QP) approach for generating smooth, C1 continuous paths for mobile robots using piece-wise quadratic Bezier (PWB) curves. Our method explicitly incorporates safety margins within a structured optimization framework, balancing trajectory smoothness and robustness with manageable numerical complexity suitable for real-time and embedded applications. Comparative simulations demonstrate clear advantages over traditional piece-wise linear (PWL) path planning methods, showing reduced trajectory deviations, enhanced robustness, and improved overall path quality. These benefits are validated through simulations using a Pure-Pursuit controller in representative scenarios, highlighting the practical effectiveness and scalability of our approach for safe navigation.
Control software components in the Cyber-Physical Systems domain use Function Blocks (FBs) to implement system logic, which manages interactions among hardware components (e.g., sensors and actuators) and other control software modules. FBs provide discrete input and output interfaces that support unidirectional data and event connections, which facilitate the integration of multiple subsystems to meet overall system requirements. However, this design based on discrete interfaces results in a high number of connection instances and lacks encapsulation. The complexity is further compounded by implicit one-to-many (fan-out) connections among subsystems, which can hinder both system understandability and maintainability. To address these challenges, we propose leveraging the adapter interface concept from IEC 61499. This approach replaces multiple discrete connections with a single bidirectional adapter connection that explicitly defines interaction requirements between components. Consequently, this reduces visual complexity and enhances system maintainability and testability.
The developer community has made significant efforts to enable fast and resilient implementations of industrial applications centred around the Asset Administration Shell (AAS). Software Developer Kits (SDKs) help industry and academia to streamline the development process and reduce errors when working with the AAS metamodel or serving the AAS API. In this work, we analyze established SDKs and other supporting tools that facilitate the development of AAS SDKs and applications. We introduce four workflows with varying degrees of automation that leverage these tools, discussing both current and anticipated challenges related to code generation, along with effective mitigation strategies. These workflows are qualitatively compared regarding their implementation and management effort, required skills, correctness, as well as future-proof maintainability. Based on six study cases of real-world AAS implementations, we ascertain the extent to which these workflows are already being applied in practice.
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.
The debugging capabilities in the automation domain lag behind those in other domains. One obstacle to debugging systems in the automation domain is the presence of heavy and costly machinery connected to the Programmable Logic Controllers (PLCs). Replay debugging presents an opportunity to overcome this issue by allowing the user to debug a system with full control and observability, without the real system being connected. During the first stage of replay debugging, information is captured while the system is running. This is then used during the second stage to reproduce the system in a safe environment.This work focuses on a viability analysis of replay debugging for Function Block networks within resources modelled according to the IEC 61499 standard. In order to reduce the execution impact of information capture, the work presents an approach where only a small subset of the run-time information is captured and can be used in an offline stage to generate the full execution behaviour of an IEC 61499 resource. Thorough tests and experiments show that the reproduction of the original states is possible.
Vibration monitoring is one of the most widely used approaches to detect infrastructural deterioration. This work presents an ongoing project focused on vibration-based structural health monitoring (SHM) of bridges using dense sensor networks and artificial neural networks (ANNs). A key challenge in this context is to manage the large volumes of high-frequency data generated by distributed triaxial accelerometers deployed across bridge spans. To address this, two data-centric contributions are proposed and evaluated: (1) a lightweight, threshold-based event filtering method that selectively retains high-intensity vibration segments while discarding low-amplitude non-informative vibrations, and (2) the evaluation of the Parquet data format, which significantly improves storage efficiency and compatibility with scalable analytical pipelines. Experimental results demonstrate that the proposed filtering method can achieve substantial data storage reduction retaining as little as 20–40% of the original data while maintaining low signal distortion.
The rising number of truck-related fatalities on Brazilian highways, alongside global increases in heavy-duty vehicle (HDV) incidents, highlights the need for advanced adaptive safety systems. Responsibility-Sensitive Safety (RSS) provides a verifiable framework for autonomous driving but struggles with HDV challenges such as highly variable mass. This work proposes TruckRSS, an RSS extension for variable-mass vehicles, integrating mass estimation through Neural Networks (NN) and formal runtime verification using Signal Temporal Logic (STL). A data-driven approach employing Gated Recurrent Unit (GRU) models achieved a Mean Absolute Error (MAE) of 121.91 kg after post-processing. Experimental validation in the CARLA simulator demonstrated that TruckRSS adapts braking behavior based on real-time mass estimation, reducing unnecessary following distances and improving traffic flow efficiency compared to standard RSS. Real-time monitoring evaluations confirmed that TruckRSS can be enforced with minimal computational overhead, supporting its scalability for deployment in safety-critical HDV applications.
This paper presents an enhanced fault-tolerant architecture for industrial Networked Control Systems (NCS) that simultaneously supports supervision video transmission and maintains strict real-time constraints. Building upon previous work that optimized video and control data flows through static traffic management techniques, this study introduces approaches to address fault tolerance and scalability challenges. First, supervision video enhanced quality operating up to the standard 50 frames per second (fps) was tested. Second the use of Field Programmable Gate Array (FPGA) resources embedded within the core network switch to handle controller fault recovery locally is proposed and tested. The use of this FPGA-enhanced architecture effectively isolates control and video traffic, ensuring real-time responsiveness even with supervisory cameras operating at high frame rates. Simulation results confirm maximum delays in faulty scenarios, well within the real-time requirements. The proposed solutions demonstrate possible scalable methods for integrating fault tolerance and high-quality video supervision in modern industrial NCS, with the FPGA-based design offering latency improvements of 7.58% over previously studied designs.
Vibration is one of the leading causes of unplanned shutdowns, hydrocarbon leaks, and industrial accidents, particularly due to fatigue failures. Traditional monitoring methods require direct contact with the structure, which is not always feasible. However, computer vision-based approaches have emerged in recent years as promising alternatives for non-contact vibration analysis. This study proposes a method for measuring vibration in pipelines using high-speed cameras and motion amplification techniques based on optical flow tracking. The approach estimates displacement by tracking key feature points across video frames using the Lucas–Kanade method and averaging their trajectories. The dominant vibration frequency is determined by Fourier analysis of the mean displacement signal. A bandpass filter centered around this frequency is then applied to isolate the relevant motion components. Based on this analysis, the dominant frequency component is amplified in the resulting video. For a typical setup with a distance of 30 cm between the camera and the pipeline, using a camera with a field of view of 54° and a vertical resolution of 720 pixels, the minimum perceptible displacement is approximately 0.02 mm. Similarly, the maximum measurable frequency is constrained by the sampling rate of the camera. The incorporation of computer vision techniques for non-contact vibration analysis expands available tools for predictive fault monitoring, contributing to industrial maintenance and accident prevention.
In this paper, the use of Large Language Models (LLMs) for the configuration of hybrid TSN/5G networks is investigated. We discuss promising use cases where LLMs offer significant potential to simplify complex configuration tasks. Particularly, we consider two important scenarios: In the first scenario, the LLM functions as an engineering component enhancing traditional network control entities such as the Centralized User Configurations (CUCs) and Centralized Network Configurations (CNCs) of TSN networks and interacting with the 5G control plane. In the second scenario, the LLM serves as an interactive assistance tool for users performing manual configuration tasks. For these use cases, an LLM-based architecture for network configuration is proposed, which consists of a Retrieval-Augmented Generation (RAG) system, a verification component, and an orchestration layer. Within the framework of this architecture, we introduce LLM-based methods to enhance the reliability of configuration in complex real-time networks, leveraging strategies such as divide and conquer, prompt engineering, and verification.
LoRaWAN supports bi-directional communication, allowing end-devices not only to transmit data to the network server (uplink), but also to receive commands or configuration updates from the network server (downlink). Although most of the research has focused on uplink performance, downlink traffic plays a critical role in enabling control, configuration, and actuation in industrial scenarios. LoRaWAN defines three device classes that offer different Quality of Service (QoS) levels for downlink communication, but they do not specifically address reliability. This work introduces three QoS classes for LoRaWAN downlink that enable configurable trade-offs between reliability and end-device energy consumption. Moreover, this work investigates novel gateway selection strategies that dynamically schedule downlink traffic according to the required QoS levels. Preliminary simulation results show that the proposed strategy significantly improves the packet delivery ratio for downlink transmissions requiring reliability, especially in scenarios with mixed QoS requirements, at the expense of increased energy consumption.
The convergence of IT and OT networks introduces strict latency and security requirements, especially in virtualized industrial environments. While TSN and DetNet provide bounded-latency traffic delivery, traditional software firewalls break determinism due to variable rule evaluation times. Hence, we propose a time-limited firewall design that limits per-packet rule evaluation time to a fixed budget, ensuring deterministic processing even under high load. To preserve security despite partial rule checks, we propose that a deferred filtering stage verifies and, if necessary, retroactively corrects earlier forwarding decisions. We implement this design in a software firewall prototype and evaluate it under maximum packet rate. The system guarantees limited latency for all packets, maintains high throughput, and ensures eventual security consistency, all without requiring specialized hardware.
In the automotive context, the quality of products or assemblies is inspected regularly at quality gates, constituting a laborious and error-prone process if performed manually. Machine Vision (MV) systems can partially automate the inspection processes, relieving workers from such tasks and thus increasing the overall productivity. Although MV systems offer high potential in theory, their performance depends on large volumes of annotated data specific to the considered use cases. Yet, acquiring and annotating such data in industrial settings poses significant challenges due to accessibility constraints, confidentiality issues, high manual effort, and costs. As an alternative to the acquisition of real-world data, this work investigates the utilization of synthetic data for the training of deep learning (DL) models for defect detection, providing high-quality annotations by design and reducing manual effort significantly. The authors present a methodology for image generation and demonstrate that a DL model trained on synthetic data achieves comparable accuracy in object detection to one trained on real-world data. Moreover, the presented approach offers the capability to generate explicit error cases, providing valuable insights for error mitigation strategies in the context of automotive production and assembly processes.
