10784 research outputs found
Sort by
Explaining 3D Semantic Segmentation Through Generative AI-Based Counterfactuals
Get PDFPublisher Copyright: © 2025 The Author(s). Expert Systems published by John Wiley & Sons Ltd.Interpreting the predictions of deep learning models on 3D point cloud data is an important challenge for safety-critical domains such as autonomous driving, robotics and geospatial analysis. Existing counterfactual explainability methods often struggle with the sparsity and unordered nature of 3D point clouds. To address this, we introduce a generative framework for counterfactual explanations in 3D semantic segmentation models. Our approach leverages autoencoder-based latent representations, combined with UMAP embeddings and Delaunay triangulation, to construct a graph that enables geodesic path search between semantic classes. Candidate counterfactuals are generated by interpolating latent vectors along these paths and decoding into plausible point clouds, while semantic plausibility is guided by the predictions of a 3D semantic segmentation model. We evaluate the framework on ShapeNet objects, demonstrating that semantically related classes yield realistic counterfactuals with minimal geometric change, whereas unrelated classes expose sharp decision boundaries and reduced plausibility. Quantitative results confirm that the method balances defined interpretability metrics, producing counterfactuals that are both interpretable and geometrically consistent. Overall, our work demonstrates that generative counterfactuals in latent space provide a promising alternative to input-level perturbations.Peer reviewe
Dynamic Characterization of the Consistory Building in Comune di Sant’Agapito, Italy: Ambient Vibration Testing, Data Analysis, and Finite Element Validation
No full textPublisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.This research paper presents a comprehensive study on the dynamic behaviour of the consistory building in Comune di Sant’Agapito, Italy, through ambient vibration testing and subsequent data analysis. The consistory building, a significant historical structure, was subjected to an ambient vibration test to capture its natural frequencies and mode shapes. Accelerometers were strategically placed to record the building’s response to ambient vibrations. The collected data was then processed using two distinct methods: Fast Fourier Transform (FFT) and Frequency Domain Decomposition (FDD). FFT and FDD were chosen for their complementary strengths in identifying frequency content and extracting modal parameters, respectively. The differences between the results obtained from FFT and FDD were meticulously analysed to understand the advantages and limitations of each method in the context of historical buildings. To validate the experimental findings, a finite element model (FEM) of the consistory building is developed to simulate its dynamic response. The FEM results were compared with the experimental data to assess the accuracy of the model and the reliability of the FFT and FDD methods. This study not only contributes to the understanding of the consistory building’s dynamic characteristics but also demonstrates the efficacy of combining ambient vibration testing, advanced data processing techniques, and finite element modelling.Peer reviewe
Using IoT Technologies to Facilitate Human – Machine Communication: A Use Case for Setup Time Acquisition
No full textPublisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.Industry 4.0 is changing the industrial paradigm. The combination of cyber-physical systems (CPS) with 4.0 technologies has endless possibilities to im-prove the industrial environment. This article focuses on the combination of CPS and the Internet of Things (IoT) to facilitate communication between CPS and hu-mans, a topic in which research is still lacking. In particular, this research aims to overcome the challenge of accurately knowing the setup time of machine tools. The machine tool's sensing system is not able to measure the setup time because it is switched off during this process. However, the incorporation of an additional device can solve this problem. This paper presents a case study that uses an IoT device with Radio Frequency Identification (RFID), so that the user can communicate with the CPS and know what the machine setup time has been, in order to calculate the Overall Equipment Effectiveness (OEE) considering or ignoring the setup time. The results conclude that IoT improves the communication between the CPS and the user and shows the importance of setup time to correctly measure productivity and define improvement strategies.Peer reviewe
Evaluating reinforcement learning-based neural controllers for quadcopter navigation in windy conditions
Get PDFPublisher Copyright: © 2025 The AuthorsAccurate quadcopter navigation under windy conditions remains challenging for traditional control methods, especially in the presence of unpredictable wind gusts and strict navigational constraints. This paper evaluates Deep Reinforcement Learning (DRL) based controllers under such conditions, analysing the impact of wind domain randomisation, multi-goal training, enhanced state representations with explicit wind information, and the use of temporal data to capture affecting dynamics over time. Experiments in the AirSim simulator across four trajectories — evaluated under both no-wind and windy conditions — demonstrate that DRL-based controllers outperform classical methods, particularly under stochastic wind disturbances. Moreover, we show that training a DRL agent with domain randomisation improves robustness against wind but reduces efficiency in no-wind scenarios. However, incorporating wind information into the agent's state space enhances robustness without sacrificing performance in wind-free settings. Furthermore, training with stricter waypoint constraints emerges as the most effective strategy, leading to precise trajectories and improved generalisation to wind disturbances. To further interpret the learned policies, we apply Shapley Additive explanations analysis, revealing how different training configurations influence the agent's feature importance. These findings underscore the potential of DRL-based neural controllers for resilient autonomous aerial systems, highlighting the importance of structured training strategies, informed state representations, and explainability for real-world deployment.Peer reviewe
Black carbon aerosols in China: Spatial-Temporal variations and lessons from long-Term atmospheric observations
No full textPublisher Copyright: © Copyright:Black carbon (BC) significantly influences climate, air quality, and public health, and long-Term observations are essential for understanding its adverse effects. While previous studies have primarily focused on spatiotemporal variations, deeper insights from such datasets remain uncovered. Using 13 years (2008-2020) of continuous measurements of equivalent black carbon (eBC) in China, this study reported the spatial-Temporal variations of eBC and its sources, including solid fuel (eBCsf) and liquid fuel combustion (eBClf). The results showed that eBC and its sources exhibited higher concentrations in eastern and northern China compared to western and southern China. Seasonal variations of eBC and eBCsf generally showed lower values during summer and higher values during winter at most stations. Long-Term trends indicated that eBC and eBClf decreased most rapidly at urban stations, while eBCsf declined faster at rural stations. Comparisons of eBC concentrations and trends between this study and global observations revealed higher eBC levels but lower reduction rates in China. These long-Term observations showed that the model simulations performed well in simulating spatial distribution but poorly in capturing inter-Annual variations. The weather-normalized eBC concentrations showed potential for adjusting emission estimates. The normalized results also suggested that emission control was the dominant driver of the BC reduction. This decrease was primarily driven by reductions from solid fuel combustion at rural and background stations. This study provides insights for reducing uncertainties in black carbon emission inventories and improving model performance in simulating surface concentrations.Peer reviewe
Mechanical power is not associated with mortality in COVID-19 mechanically ventilated patients
Get PDFPublisher Copyright: © The Author(s) 2025.Background: The relative contribution of the different components of mechanical power to mortality is a subject of debate and has not been studied in COVID-19. The aim of this study is to evaluate both the total and the relative impact of each of the components of mechanical power on mortality in a well-characterized cohort of patients with COVID-19-induced acute respiratory failure undergoing invasive mechanical ventilation. This is a secondary analysis of the CIBERESUCICOVID project, a multicenter observational cohort study including fifty Spanish intensive care units that included COVID-19 mechanically ventilated patients between February 2020 and December 2021. We examined the association between mechanical power and its components (elastic static, elastic dynamic, total elastic and resistive power) with 90-day mortality after adjusting for confounders in seven hundred ninety-nine patients with COVID-19-induced respiratory failure undergoing invasive mechanical ventilation. Results: At the initiation of mechanical ventilation, the PaO2/FiO2 ratio was 106 (78; 150), ventilatory ratio was 1.69 (1.40; 2.05), and respiratory system compliance was 35.7 (29.2; 44.5) ml/cmH2O. Mechanical power at the initiation of mechanical ventilation was 24.3 (18.9; 29.6) J/min, showing no significant changes after three days. In multivariable regression analyses, mechanical power and its components were not associated with 90-day mortality at the start of mechanical ventilation. After three days, total elastic and elastic static power were associated with higher 90-day mortality, but this relationship was also found for positive end-expiratory pressure. Conclusions: Neither mechanical power nor its components were independently associated with mortality in COVID-19-induced acute respiratory failure at the start of MV. Nevertheless, after three days, static elastic power and total elastic power were associated with lower odds of survival. Positive end-expiratory pressure and plateau pressure, however, captured this risk in a similar manner.Peer reviewe
Leveraging constraint programming in a deep learning approach for dynamically solving the flexible job-shop scheduling problem
Get PDFPublisher Copyright: © 2024 The AuthorsRecent advancements in the flexible job-shop scheduling problem (FJSSP) are primarily based on deep reinforcement learning (DRL) due to its ability to generate high-quality, real-time solutions. However, many DRL approaches struggle with large action spaces, making the search for optimal solutions computationally intensive and impacting performance. Moreover, established techniques like exact methods and constraint programming (CP) often achieve better results for smaller instances. This paper aims to fully harness the strengths of these existing techniques by integrating CP within a deep learning (DL)-based methodology, leveraging the benefits of both. In this paper, we introduce a method that involves training a DL model using optimal solutions generated by CP, ensuring the model learns from high-quality data, thereby eliminating the need for the extensive exploration typical in DRL and enhancing overall performance. Further, we integrate CP into our DL framework to jointly construct solutions, utilizing DL for the initial complex stages and transitioning to CP for optimal resolution as the problem is simplified. Our hybrid approach has been extensively tested on three public FJSSP benchmarks, demonstrating superior performance over five state-of-the-art DRL approaches and a widely-used CP solver. Additionally, with the objective of exploring the application to other combinatorial optimization problems, promising preliminary results are presented on applying our hybrid approach to the traveling salesman problem, combining an exact method with a well-known DRL method.Peer reviewe
Overlap Number of Balls Model-Agnostic CounterFactuals (ONB-MACF): A data-morphology-based counterfactual generation method for trustworthy artificial intelligence
No full textPublisher Copyright: © 2025 Elsevier Inc.Explainable Artificial Intelligence (XAI) is a pivotal research domain aimed at clarifying AI systems, particularly those considered “black boxes” due to their complex, opaque nature. XAI seeks to make these AI systems more understandable and trustworthy, providing insight into their decision-making processes. By producing clear and comprehensible explanations, XAI enables users, practitioners, and stakeholders to trust a model's decisions. This work analyses the value of data morphology strategies in generating counterfactual explanations. It introduces the Overlap Number of Balls Model-Agnostic CounterFactuals (ONB-MACF) method, a model-agnostic counterfactual generator that leverages data morphology to estimate a model's decision boundaries. The ONB-MACF method constructs hyperspheres in the data space whose covered points share a class, mapping the decision boundary. Counterfactuals are then generated by incrementally adjusting an instance's attributes towards the nearest alternate-class hypersphere, crossing the decision boundary with minimal modifications. By design, the ONB-MACF method generates feasible and sparse counterfactuals that follow the data distribution. Our comprehensive benchmark from a double perspective (quantitative and qualitative) shows that the ONB-MACF method outperforms existing state-of-the-art counterfactual generation methods across multiple quality metrics on diverse tabular datasets. This supports our hypothesis, showcasing the potential of data-morphology-based explainability strategies for trustworthy AI.Peer reviewe
A comprehensive survey of Federated Intrusion Detection Systems: Techniques, challenges and solutions
Get PDFPublisher Copyright: © 2024 The AuthorsCyberattacks have increased radically over the last years, while the exploitation of Artificial Intelligence (AI) leads to the implementation of even smarter attacks which subsequently require solutions that will efficiently confront them. This need is indulged by incorporating Federated Intrusion Detection Systems (FIDS), which have been widely employed in multiple scenarios involving communication in cyber–physical systems. These include, but are not limited to, the Internet of Things (IoT) devices, Industrial IoT (IIoT), healthcare systems (Internet of Medical Things/IoMT), Internet of Vehicles (IoV), Smart Manufacturing (SM), Supervisory Control and Data Acquisition (SCADA) systems, Multi-access Edge Computing (MEC) devices, among others. Tackling the challenge of cyberthreats in all the aforementioned scenarios is of utmost importance for assuring the safety and continuous functionality of the operations, crucial for maintaining proper procedures in all Critical Infrastructures (CIs). For this purpose, pertinent knowledge of the current status in state-of-the-art (SOTA) federated intrusion detection methods is mandatory, towards encompassing while simultaneously evolving them in order to timely detect and mitigate cyberattack incidents. In this study, we address this challenge and provide the readers with an overview of FL implementations regarding Intrusion Detection in several CIs. Additionally, the distinct communication protocols, attack types and datasets utilized are thoroughly discussed. Finally, the latest Machine Learning (ML) and Deep Learning (DL) frameworks and libraries to implement such methods are also provided.Peer reviewe
Enhancing Mass Transport in Organic Redox Flow Batteries Through Electrode Obstacle Design
Get PDFPublisher Copyright: © 2025 by the authors.This study examines the impact of incorporating obstacles in the electrode structure of an organic redox flow battery with a flow-through configuration. Two configurations were compared: A control case without obstacles (Case 1) and a modified design with obstacles to enhance mass transport and uniformity (Case 2). While Case 1 exhibited marginally higher discharge voltages (average difference of 0.18%) due to reduced hydraulic resistance and lower Ohmic losses, Case 2 demonstrated significant improvements in concentration uniformity, particularly at low state-of-charge (SOC) levels. The obstacle design mitigated local depletion of active species, thereby enhancing limiting current density and improving minimum concentration values across the studied SOC range. However, the introduction of obstacles increased flow resistance and pressure drops, indicating a trade-off between electrochemical performance and pumping energy requirements. Notably, Case 2 performed better at lower flow rates, showcasing its potential to optimize efficiency under varying operating conditions. At higher flow rates, the advantages of Case 2 diminished but remained evident, with better concentration uniformity, higher minimum concentration values, and a 1% average increase in limiting current density. Future research should focus on optimizing obstacle geometry and positioning to further enhance performance.Peer reviewe