Contamination diagnosis in water distribution systems (WDS) is critical for ensuring public health and maintaining water quality. However, exact localization of contamination sources is often highly challenging due to sparse sensor deployment, uncertain water flow dynamics, and the ambiguity of contamination signatures. To address this challenge, we propose a novel end-to-end deep learning framework for ranked contamination diagnosis in WDS. Instead of relying on a single-source prediction, the deep learning model outputs a probabilistic ranking of likely contamination source nodes, enabling top- $k$ diagnosis strategies under uncertainty. A unified classification layer jointly performs contamination detection and source localization, allowing the model to operate in an end-to-end fashion. Extensive experiments on a benchmark WDS demonstrate that the proposed framework achieves strong top- $k$ diagnosis accuracy and remains robust across varying sensor configurations. This work offers a practical and scalable approach to contamination diagnosis in WDS, supporting more informed and effective decision-making in real-world operations. The code is available at: https://github.com/Xiaohan-Chen/ST-GAT.
This paper presents a benchmark for fault detection and isolation (FDI) in floating offshore wind farms, addressing the lack of standardized evaluation frameworks for this emerging technology. Developed using the FOWLTY simulator, the benchmark models a wind farm with seven floating turbines based on the NREL 5MW reference turbine and DeepCWind platform. It incorporates ten diverse wind scenarios (5-23 m/s) and realistic fault conditions, including sensor and actuator faults with variable severity and timing. The benchmark provides sensor measurements with injected noise, actuator reference signals, and evaluation metrics such as false alarm rate (FAR), missed detection rate (MDR), and correct isolation rate. By offering a modular Simulink environment and customizable datasets, this work enables reproducible comparisons of FDI methods while highlighting the unique dynamics of floating offshore systems. The benchmark is publicly available to support research in fault detection and fault-tolerant control for offshore wind energy.
The EvoRoads project develops a connectivity platform that digitalises transport infrastructure and integrates safety assessment services. Leveraging advanced custom-made AI models, EvoRoads analyses infrastructure monitoring data at various geospatial levels, enabling proactive risk warnings and supporting road operators in enhancing safety and operational efficiency. It also defines safety criteria and key performance indicators (KPIs) within the “Safe System” approach. By focusing on data-driven methods, EvoRoads enhances decision making in transportation systems, ensuring the design for reliability and safety while improving maintenance policies. The platform's ability to integrate real-time insights fosters smarter, more responsive infrastructure management, contributing to more efficient and resilient road networks. Co-created with local stakeholders, the EvoRoads use cases are grounded in real-world needs, enhancing the relevance and impact of the solutions. EvoRoads also sets a precedent for future-proof, collaborative urban mobility, advancing the European Vision Zero initiative and providing a blueprint for cities across Europe to enhance their road safety systems. By fostering data-sharing and encouraging transparency, EvoRoads champions an open-data ethos that not only improves safety outcomes but also drives innovation within the broader mobility ecosystem.
The abrupt growth of the world population has raised global concerns about sustainable energy supply. In response, energy policies are continuously evolving to guarantee stability across the energy sector. Stakeholders in the energy supply chain are striving to maintain reliability and balance in the system. Renewable energy sources, because of their low generation cost and sustainability, are increasingly replacing fossil fuel-based power plants. However, their integration introduces significant uncertainty and variability. In this context, accurate forecasting becomes essential for effective planning in renewable-powered microgrids. Reliable forecasts enhance the availability, stability, and overall reliability of microgrids, thereby supporting a resilient and efficient energy supply chain. To address this need, a novel bi-long short-term memory attention algorithm is proposed to enhance the forecasting accuracy of key microgrid parameters that drive power-system decision making. The proposed model demonstrates superior performance over an existing electrical net-load forecasting approach that combines deep neural networks with wavelet transform on the same real-world dataset. Specifically, it achieves a 4% reduction in Mean Absolute Percentage Error and a 2% decrease in Root Mean Square Error. Furthermore, the proposed model accurately tracks the peak and trough points in photovoltaic and wind energy, which is a key factor in forecasting.
This paper presents the design of an interval observer (IO) for systems characterized by nonlinearities that are assumed to fulfill one-sided Lipschitz quadratically innerbounded (OSL-QIB) conditions. This novel approach extends conventional design of IO for Lipschitz systems to OSL-QIB systems with the aim of improving observer design and convergence. Additionally, this new IO design is applied to fault detection, and state estimation which is crucial to guarantee normal process operation. A bioreactor for biomass production and a FHN system are considered as case studies to test the performance of the approach for fault detection and state estimation purposes, respectively.
This paper deals with the problem of detecting, isolating and estimating of multiple sensor and actuator faults in stochastic linear discrete-time systems. A linear state filter is designed to generate minimum variance output residuals having directional properties in response to sensor and actuator fault structurally detectable at current time. The filter's gain is obtained by minimizing the state prediction errors covariance matrix under decoupling constraints activated as late as possible from a feedback information on the real-time detectability conditions of sensor and actuator faults. During a finite-time structural transient, the degrees of freedom used to minimize the trace of the state prediction error covariance matrix are maximized. Two dual solutions for multiple sensor and actuator faults detection and isolation are derived, the first based on one fault isolation filter generating minimum size structured detection signals and the second based on a bank of fault isolation filter generating maximum size white structured detection signals. The convergence and stability conditions of the optimized fault isolation filter with structural transient are established from an upper bound of the state prediction errors covariance matrix obtained when the decoupling constraints are always activated.
Networked control systems that operate urban drinking water networks are susceptible to cyber-attacks while still being required to satisfy service and safety constraints. We address resilient trajectory tracking for the aggregate Barcelona network and present a dual-layer predictive control architecture. During attack-free conditions, a nominal Model Predictive Controller (MPC) regulates the plant and buffers admissible inputs. A set-membership consistency test monitors the state's evolution through families of one-step controllable sets; any violation prompts a switch to an ADMM-based Encrypted MPC solved in the cloud via homomorphic encryption, thereby preserving data confidentiality and closed-loop stability. We establish feasibility and bounded tracking errors for arbitrary switches between the two layers.
This paper proposes a non-invasive method for the detection of parameter variation in DC-DC buck power converters. It is computationally efficient and only requires the usual current and voltage sensors. The method is presented in several steps. First, a generic buck converter model is presented. Then, the general design of a sliding mode observer is described as well as its method of fault reconstruction. This sliding mode observer is then adapted to detect parameter changes of the buck converter model. Finally, simulations are performed to validate the observer design.
Water utilities around the world typically use chlorine as the main disinfectant for ensuring high-quality drinking water. Usually, a few fixed sensors monitor water quality by detecting changes in parameters like chlorine residuals, guiding chlorination strategies. However, limited sensor coverage leaves most parts of the network unmonitored. Additionally, rapid urban growth and climate change complicate water quality dynamics, challenging conventional methods for sparse to dense state estimation. In this work, we propose a neural network based surrogate model for efficiently obtaining time-dependent approximations of the chlorine concentration dynamics in a water distribution system. We incorporate this surrogate model into an extended Kalman filter to estimate all chlorine concentration states on the basis of only a few sensors. We perform extensive empirical evaluations on popular benchmark water distribution systems from the literature.
This paper presents a hybrid fault diagnosis approach for wind turbines that integrates structural analysis through Analytical Redundancy Relations (ARRs) with datadriven modeling using Gaussian Process Regression (GPR). The proposed method leverages the physical structure of the system to define input-output dependencies and trains GPR estimators on fault-free operational data to predict key subsystem outputs. Residuals are computed by comparing sensor measurements with GPR predictions, and faults are detected using a combination of interval-based thresholds and Cumulative Sum (CUSUM) control charts. The proposed approach is validated on a simulated 5-MW wind turbine benchmark model under realistic operating conditions. Various fault scenarios are injected in the pitch actuator, drivetrain, and generator subsystems. Results demonstrate the fault diagnosis accuracy, robustness, and early detection capability across diverse fault types.
Self-healing control (SHC) plays a critical role in ensuring the stable operation of complex industrial processes, particularly in systems like the fused magnesia furnace (FMF), where fluctuations in operational conditions can lead to significant disruptions. Unlike fault-tolerant control, which maintains basic system functionality under fault conditions, SHC incorporates intelligent optimization to enable adaptive adjustment and dynamic recovery. To this end, this paper proposes data-rule integrated adaptive fuzzy neural networks for SHC of the FMF. First, an attention-driven case-based reasoning (CBR) approach is proposed to dynamically analyze and match historical operational data, accurately identifying cases most relevant to the current operating conditions. Second, an adaptive fuzzy neural network incorporating expert rules is constructed as a compensation module, which optimizes and adjusts the CBR results in real-time. Experimental results demonstrate that the proposed method exhibits significant advantages in handling operational condition changes and data drift, achieving notable improvements in control accuracy and response speed compared to traditional CBR method.
This paper aims at demonstrating how the dwell-time fault-tolerant control (FTC) approach can be used to accommodate faults occurring in control surfaces of high speed aircraft. The application support is the F-8 aircraft that performs a coordinated turn. Global exponential stability of the FTC law is proven, considering the coupling between the fault diagnosis and control units. Beyond the application case, the paper demonstrates how the structured H-infinity approach can be embedded within the dwell-time FTC theory. A simulation campaign considering sensor noise, control surfaces saturation and uncertainties in mass, inertia and aerodynamic coefficients, demonstrates the potential of the proposed approach.
In this paper, a data-driven Fault Detection strategy is proposed for detecting faults in generator speed sensors of floating offshore wind turbines using Machine Learning (ML) techniques. To generate reliable datasets in both healthy and faulty conditions, a seven-turbine offshore wind farm is set up using the FOWLTY MATLAB/Simulink benchmark framework. A systematic approach is applied for signal-based feature extraction, followed by dimensionality reduction through feature selection using LightGBM. Two classifiers-Support Vector Machines (SVM) and Decision Trees (DT)-are trained and evaluated on the resulting datasets. The results show that both methods achieve acceptable accuracy in fault classification, but the SVM demonstrates better performance compared to the DT. These findings confirm that the proposed methods can accurately detect generator speed sensor faults in offshore wind turbines. The results confirm that the proposed methods can accurately detect generator speed sensor faults in offshore wind turbines.
Biofilms in drinking water distribution systems (DWDS) pose a critical challenge to water quality. If left unchecked, they can compromise the biological stability of delivered water and ultimately public health. Existing biofilm sensing techniques primarily focus on metabolic or genetic indicators of activity, often using local and destructive methods. While rich in information, such data are difficult to apply in developing practical biofilm growth models. Biofilm thickness, however, is a more representative and scalable metric for this purpose. Yet, limited research exists on non-invasive thickness sensing in DWDS. This study introduces two non-destructive methods for measuring biofilm thickness by leveraging changes in heat resistance and residence time. Heat resistance was evaluated using ambient and water temperature measurements, while residence time was assessed with a conservative tracer. Both techniques were tested in the Slimer experimental setup (50 m long, 13.2 mm diameter PVCp pipe) under realistic hydraulic conditions. Results showed a strong correlation between biofilm thickness and residence time drift, indicating flow disturbance as a reliable indicator of biofouling. In contrast, heat resistance sensing exhibited considerable natural variability, limiting its analytic value. The findings highlight residence time analysis as a promising, non-invasive approach for estimating biofilm thickness. This method offers continuous, non-destructive monitoring, enabling early detection of biofilm-related anomalies and providing valuable input for both laboratory and field applications aimed at enhancing DWDS resilience.
This paper addresses the modeling and control of Diesel engine injection systems, focusing on a control strategy based on fuzzy static output feedback. The main objective is to regulate the pump and rail pressures toward desired reference values by manipulating both the engine speed and the injector control signals. The system is subject to external disturbances assumed to be bounded within a specific frequency range. To accurately describe the system's nonlinear behavior, a Takagi-Sugeno (TS) fuzzy model is employed. Based on this representation and a descriptor formulation, a static output feedback controller is designed to ensure pressure regulation and satisfy performance specifications formulated as linear matrix inequalities (LMIs). The effectiveness and robustness of the proposed control scheme are demonstrated through detailed simulation studies.
The health monitoring of key processes or subsystems in the manufacturing industry is of capital importance to improve maintenance policies. This work will look for degradation patterns in the aluminum die casting process used to generate hybrid engine blocks in the automotive industry. The die-casting process that generates each engine block is rather complex, and each die-casting machine can generate a new engine block every 90 seconds. Within this process, the aluminum injection stage is critical and it lasts only few seconds. The injection device monitoring system provides 2000 measurements of several physical variables involved in the process. This work has faced the challenges of providing an estimation of the degradation patterns of the piston head for one of these injection machines in a factory, using those time series obtained from the controller, and also to identify unusual malfunction patterns found during a three-month period. The problem was tackled using both traditional and deep machinelearning techniques for unsupervised learning. Results have shown that degradation patterns can be identified, but the presence of unexpected malfunctions will require additional techniques and/or data.
This paper presents novel findings on the application of Individual Pitch Control (IPC) for multi-megawatt, two-bladed downwind turbines, as demonstrated in a Goldwind turbine design. The results show that IPC effectively reduces fatigue loads by 10-20% and extreme loads by 5-10%. This is achieved through optimization strategies that utilize rotor lift and activation techniques to balance load reduction with the lifetimes of the main and blade bearings, meaning the IPC algorithm have been tuned for different objectives. Furthermore, the study explores a passive yaw configuration, where yaw control is achieved via IPC (yaw-by-IPC), demonstrating a significant 45% reduction in tower top/yaw torsional moment loads while maintaining load increases on other components below 5%. This approach supports a more cost-efficient tower design, enabling the use of a lattice tower structure.
This paper introduces an interpretable ensemble machine learning approach specifically designed for fault detection in floating offshore wind turbines. The methodology integrates advanced statistical features extracted from residual signals with complementary machine learning models, enhancing the identification of subtle fault-induced deviations typical in offshore environments. Validated using a realistic offshore wind farm simulation benchmark, the proposed method demonstrated clear advantages over traditional threshold-based techniques and single-model approaches. The practical interpretability of the method is demonstrated through analysis of feature relevance, aiding effective fault diagnosis. Although tested primarily on specific sensor faults, the modular nature of the methodology supports its generalisation and highlights its potential suitability for broader fault detection scenarios and real-time applications.
Solid Oxide Fuel Cells are promising power generation technologies, especially for large-scale applications. As the marine industry is targeting a full de-carbonization by year 2050, increasing attention is being directed toward the implementation of these technologies. Solid Oxide Fuel Cells are complex systems where thermodynamics and electrochemical reactions are coupled, resulting in highly non-linear dynamics, tight operational constraints, and multiple distributed sensors. Those quantities that cannot be directly measured, need to be estimated. Among these, the so called Area Specific Resistance is an indicator of cell's health condition, related to the cell degradation. This paper proposes a Moving Horizon Estimator based on an extended state-space model of a methane-fueled Solid Oxide Fuel Cell, to estimate in real time the Area Specific Resistance of the cell. Using the estimated value, along with its maximum and average rates, a predictive framework is developed to estimate the Remaining Useful Life of the cell. Simulations are used to illustrate the application and the efficiency of the proposed method.
This paper considers the optimal fault-tolerant control (FTC) issue for network nonlinear systems developed by an inverse nested differential game (INDG), which is composed of zero-sum games between controllers and faults in subsystems and a graphical game between subsystems. The proposed INDG-based optimal FTC framework guides the design of meaningful performance indexes with respect to inner faults and coupling faults, such that the fault-tolerant controllers can be constructed to achieve the optimality and stability of the whole system. Sufficient criteria of INDG-based optimal FTC are proposed, and the stability of the network system is rigorously proved under the designed performance indexes and optimal FTC.
