Engineering

Abstract: The increasing integration of renewable energy sources into industrial electrical systems requires effective energy management strategies to improve renewable energy utilization, reduce grid dependence, and control energy costs. This study presents a comparative analysis of two energy management strategies for a grid-connected hybrid photovoltaic (PV)–wind energy system supplying the electrical demand of an industrial plant. The hybrid system is modeled in MATLAB/Simulink, including the PV and wind subsystems and their associated maximum power extraction controllers. Two management algorithms are investigated: an optimistic strategy based on renewable power production, load priorities, and variable electricity tariffs, and an intelligent strategy based on fuzzy logic for adaptive energy flow management. Both strategies determine the power exchanges among the renewable sources, industrial load, and utility grid under varying operating conditions. Their performances are evaluated and compared in terms of renewable energy utilization, load supply, grid energy exchange, and operating cost. The simulation results demonstrate the effectiveness of both strategies in coordinating energy flows within the hybrid system, while revealing differences in their ability to utilize renewable generation and manage grid exchanges. The comparative analysis provides useful insights into the application of optimistic and fuzzy logic-based strategies for energy management in grid-connected hybrid PV–wind systems supplying industrial loads.

Abstract: Achieving high usable capacity in lithium-ion batteries is valuable only when cathode-related structural and electrode/electrolyte instability can be limited during cycling. This study develops a two-stage TabNet-XGBoost framework to predict capacity retention and cycle life from a cathode-centered, cell-contextual descriptor set spanning cathode composition and crystal structure, anode and electrolyte variables, initial electrochemical characteristics, and cycling conditions. TabNet generates 64-dimensional latent representations for XGBoost regression, while a separate descriptor-level XGBoost model supports SHAP analysis without assigning latent-feature attributions to individual physical descriptors. On a held-out test partition containing only real records, the primary model achieved R² values of 0.980 and 0.982 and RMSE values of 0.022 and 46.2 cycles for capacity retention and cycle life, respectively. A SHAP-informed particle-swarm search identified a candidate region with model-predicted capacity retention of 89.2% and cycle life of 1,419 cycles, compared with reference predictions of 84.6% and 1,185 cycles. These predictions and attributions describe associations within the represented chemistries and conditions; they do not resolve specific bulk or interfacial degradation mechanisms. The candidate is likewise a surrogate-model result that requires synthesis and controlled cycling validation. The framework therefore provides a data-driven screening tool for relating cathode structure and interface-relevant variables to cycling-stability outcomes while defining explicit boundaries for mechanistic interpretation and transfer to unseen systems.

Abstract: Assistive robots must respond appropriately when an interaction generates uncertainty, concern, or frustration, particularly in elderly-care tasks such as medication assistance. This paper proposes a control-oriented nonlinear model of human emotional dynamics for assistive human-robot interaction. The model represents happiness, sadness, anger, fear, disgust, and surprise as baseline-centred dynamic states. It combines published oscillator-based foundations for sadness and happiness with proposed appraisal inputs, emotion-to-emotion couplings, and robot-control effects. A nominal operating condition is defined, and a Jacobian linearisation is used to establish local stability of the selected numerical model. A saturated linear quadratic regulator adjusts four robot behaviours (support, predictability, task adaptation, and repair-oriented action) while respecting physical command bounds. The framework is evaluated in single and repeated unclear medication-reminder scenarios. In the repeated-reminder case, the controller reduces the positive peaks of sadness, anger, and fear by 42.00%, 36.58%, and 24.13%, respectively. A weighting study demonstrates the trade-off between stronger emotional protection and increased robot-control effort. Furthermore, under 100 parameter realisations with independent variations of up to ±20%, the fixed nominal controller reduces the mean adverse-emotion peaks in every trial and decreases the aggregate undesirable-emotion integral of absolute error by 28.36%. The results provide simulation-based evidence that the proposed model can support the analysis and bounded regulation of emotional transients in assistive interactions. Empirical calibration and human-subject evaluation remain necessary for clinical use.

Abstract: Industrial production plants in sectors such as chemicals, life sciences, metals and mining, pulp and paper, and power generation operate with a high degree of automation. During steady-state operation, control systems maintain process stability with little need for human intervention, while operators primarily supervise plant behavior and intervene during startup, shutdown, product transitions, or abnormal situations. At the same time, modern plants must monitor multiple domains simultaneously, including process performance, asset conditions, cybersecurity threats, and physical security. These developments significantly increase the number and diversity of events generated by monitoring systems. Advances in sensing technologies, connectivity, and data analytics have greatly improved the ability to detect deviations from normal operation. However, the increasing volume of detected anomalies and notifications raises the challenge of determining which events should be brought to the attention of human supervisors. Since direct human intervention becomes less frequent but more critical in highly automated plants, the effective use of operator attention becomes an essential aspect of system design. This paper examines monitoring within the broader framework of supervisory control in industrial systems. Monitoring is viewed as the continuous determination of system state across multiple domains, with anomaly detection forming the basis for subsequent supervisory decisions. Building on this perspective, the paper analyzes how detected deviations are translated into events and discusses criteria for deciding which events require human attention. A structured view of monitoring and supervision is proposed that integrates process operation, asset health, cybersecurity, and physical security. The goal is to outline design principles for event generation that support effective human oversight in increasingly autonomous industrial environments.

Abstract: This paper studies the stationary balancing of a full-scale bicycle robot using a scissored-pair control moment gyroscope (CMG). The test robot weighs 60 kg and includes two counter-rotating flywheels, a 20:1 geared gimbal system, an inertial measurement unit (IMU), and an embedded controller. For controller design, a simplified nonlinear model of the roll and CMG dynamics is derived and linearized around the upright position. The main contribution of this work is the experimental testing of a PDD2-LQR controller. This controller keeps the standard LQR state-feedback baseline and adds feedback from the roll error, roll rate, and filtered roll acceleration. Instead of redesigning the whole controller, the same LQR gains are used in both the baseline LQR and PDD2-LQR setups. This allows us to directly evaluate the benefit of the extra PDD2 terms on the same hardware under the same actuator limits. Experiments were carried out using PID, LQR, and PDD2-LQR controllers. The PDD2-LQR controller reached an average RMSE of 0.660 degrees and an average MAE of 0.477 degrees. These values represent improvements of about 51.8% and 43.2%, respectively, compared with the standard LQR controller. Overall, the results show that adding roll-motion feedback to the baseline controller significantly improves stationary balancing performance.

Abstract: This study investigates how Digital Twins, photogrammetry, and related 3D technologies are transforming cultural heritage preservation, with a particular focus on Irish case studies. Heritage sites face increasing threats from climate change, conflict, natural decay, and limited conservation resources. Traditional methods-while culturally significant-struggle to meet modern preservation demands due to their cost, time requirements, and limited precision. Digital tools offer new solutions: photogrammetry enables accurate, low-cost 3D documentation; laser scanning provides high-precision structural data; and Digital Twins support predictive maintenance, environmental monitoring, and virtual engagement. Through a mixed-methods approach combining literature review, case studies, surveys, and interviews, the research evaluates four major Irish digital heritage initiatives: Beyond 2022, Spike Island, Ogham in 3D, and the Digital Repository of Ireland. Findings show strong benefits in documentation accuracy, asset management, and public accessibility, but also highlight barriers such as funding gaps, limited digital skills, and long-term data storage challenges. The study concludes that Ireland has significant potential for leadership in digital heritage but requires a coordinated national strategy to support training, infrastructure, ethical data management, and sustainable digital preservation practices.

Abstract: Quadrotor attitude controllers are usually evaluated against wind arriving from a single direction, leaving the directional structure of attitude robustness and actuator-limited failure poorly characterized. This study compares proportional–integral–derivative (PID), linear quadratic regulator (LQR), H-infinity and sliding mode controllers within a unified framework in which a real-coded genetic algorithm and particle swarm optimization tune every configuration against an identical objective, plant and constraint set. A nonlinear six-degree-of-freedom Newton–Euler model with actuator saturation supports three scenarios: step tracking under a finite-duration wind force, sustained helical tracking, and a complete 0 to 360 degree azimuthal sweep of a body-frame disturbance moment at 15 degree resolution. Sliding mode control achieved the best helical tracking accuracy, reducing position root-mean-square error from 1.079 to 0.677 metres, the smallest directional excursion of 0.040 degrees, and the lowest saturation duty of 9.2 percent; LQR required the lowest normalized control effort; and H-infinity combined near-circular directional envelopes with nearly immediate attitude recovery. PID retained stable directional coverage over only 58.33% of azimuths, failing solely through saturation duty above the adopted limit rather than unsafe tilt. Overall, SMC and H∞ maximize disturbance suppression, whereas LQR offers the best trade-off between tracking accuracy and control efficiency for autonomous flight in windy environments.

Abstract: Semantic segmentation of food images is an essential step for the automatic analysis of meal composition. This study compares two segmentation architectures derived from U-Net and ResNet34 on the FoodSeg103 dataset, progressively evaluating the contribution of skip connections and fine-tuning. During training, U-Net improves significantly, with accuracy increasing from 44.70% to 87.70% (+43 percentage points) over 50 epochs, while ResNet34 achieves high accuracy (97.27%) from the first epoch, which then improves only marginally. For the U-Net architecture, the F1 score improves in three stages: 57.33% for the simple model, 85.11% after the addition of skip connections, and then 94.40% after fine-tuning, representing a cumulative gain of 37.1 percentage points. For ResNet34, fine-tuning yields a significantly more modest improvement, with the F1 score increasing from 62.53% to only 63.05%. This could be explained by ResNet34’s already high baseline performance, leaving less room for improvement than for U-Net. A Grad-CAM analysis is applied to ResNet34, whose particularly rapid learning from the early epochs makes it a relevant case for visually examining the regions on which the model relies for its predictions, thus complementing the quantitative evaluation with a qualitative interpretation. This experimental framework thus provides an evidence-based comparison between U-Net and ResNet34 for food image segmentation, and highlights the differentiated contribution of fine-tuning depending on the architecture considered.

Abstract: Reported accuracies for EEG-based relax-versus-stress classification are often inflated by subject-level data leakage: when epochs from the same participant appear on both sides of a train/test split, the classifier exploits subject identity as an unintended information channel. Using the public EEGMAT dataset and a multilayer perceptron trained on 63 time-domain features (mean, std, RMS across 21 channels), we contrast a leakage-permitting epoch-level split with a strict subject-grouped split, crossed with seven dataset-construction modes (full pool; Good- and Bad-Counter subsets, each raw, duplicate-balanced, or augmented) — 14 models in total. With leakage, accuracy reaches 88.2–99.2% (κ = 0.664–0.977), matching published windowed-split results. Under subject-grouped evaluation, unbalanced regimes collapse to chance or below (κ = −0.32); only balanced regimes retain genuine signal, the best reaching 85.0% accuracy, κ = 0.667, ROC-AUC = 0.996. Confusion-matrix mutual information shows leaky models resolve 36–93% of label entropy versus only 2–53% without leakage, but is sign-blind — two below-chance non-leak models still register positive mutual information — so it must be interpreted alongside κ and AUC. Cross-entropy training-loss divergence (up to 3.22 nats) offers a complementary leakage diagnostic, while weight spectral entropy and SHAP explanation entropy stay invariant across regimes, showing leakage changes what is learned, not model capacity.

Abstract: Fast and reliable detection of faults in electric motors is a critical element of properly functioning industrial systems. This paper presents a comprehensive comparative analysis of machine learning algorithms, probabilistic models and deep neural networks in the task of classifying mechanical and electrical faults in DC motors. Two different analytical approaches were proposed and verified in the study. The first, based on the extraction of structured statistical features from longer signal waveforms, was used to evaluate generative models (GNB, QDA), ensemble tree models (Random Forest, XGBoost) and Bayesian Logistic Regression. The second approach, dedicated to real-time systems, consisted of analysing raw 2-millisecond time windows using a two-channel LSTM recurrent network (analysing current and rotational speed signals) and a GNB reference model for flattened sequential data. The results of the experiments confirmed the high effectiveness of both approaches, showing the highest accuracy of XGBoost and QDA models (95%) for tabular data and LSTM networks (93%) for raw time signals. The evaluation and final selection of optimal architectures provide a foundation for further research on the rigorous quantification of epistemic and aleatoric uncertainty in industrial predictive diagnostics.

Abstract: In safety-critical applications such as industrial fault diagnosis, classification accuracy alone is not sufficient — the reliability of a model’s confidence estimates is equally important. In this work, which is a continuation of research on electric motor fault classification presented in a companion paper, pre-trained probabilistic prediction models were applied to motor state classification, and their predictive uncertainty was subsequently modelled and compared using five strategies: (1) native probabilistic outputs of single models, (2) tree level variance decomposition for Random Forest, (3) bootstrap ensemble aggregation for frequentist classifiers, (4) Markov Chain Monte Carlo (MCMC) posterior sampling for Bayesian Softmax Regression, and (5) Monte Carlo Dropout for LSTM networks. These strategies were evaluated on two complementary data representations — statistical feature vectors (56 test samples) and raw signal time windows (22372 test windows) — using six metrics: accuracy, Negative Log-Likelihood (NLL), Brier Score, Expected Calibration Error (ECE), average predictive entropy, and average epistemic variance. Additionally, an Error Detection AUC metric was introduced to assess the practical utility of uncertainty for flagging misclassifications. Taking the best results from each evaluation set: on the statistical feature set, the Bayesian Softmax model achieved the best calibration (ECE = 0.060), the lowest NLL (0.257) and Brier Score (0.157), while Random Forest attained both the highest accuracy (89.5%) and the highest Error Detection AUC (0.909). On the time series set, LSTM with MC Dropout achieved the best results across all reported metrics simultaneously: accuracy of 95.0%, NLL of 0.117, Brier Score of 0.071, near-perfect calibration (ECE = 0.015), and an Error Detection AUC of 0.945. Convergence of the MCMC sampler was confirmed by R-hat values of 1.000 for all parameters and effective sample sizes exceeding 3000. The results indicate that bootstrap ensembles consistently improve uncertainty quality across all base classifiers, and that Bayesian methods provide better calibration even when accuracy is not maximal.

Abstract: This work presents a Nonlinear Model Predictive Controller (NMPC) strategy for the optimal operation of a fuel cell system to maximize its average efficiency and minimize the degradation. Because of the different time scales associated with the different controlled variables, the proposed NMPC is divided into three distinct parts, each one customized with its specific prediction horizon and sampling time. The NMPC is based on a comprehensive model of the fuel cell stack, considering manufacturing constraints, and incorporating the compressor map. The NMPC outputs are the optimal setpoints for all the subsystems’ local controllers. The studied fuel cell system incorporates the anode recirculation loop with an ejector, a water separator and a bleed valve, and the proposed NMPC determine the optimal opening percentage of the bleed valve to prevent nitrogen accumulation and efficiency distortion. The performance of the proposed NPMC in terms of efficiency is compared with an offline setpoint generator through simulation. The results show the efficiency improvement using the proposed strategy. Furthermore, considering the unpredictable changes in load current within automotive applications, the study evaluates the efficiency when a constant load is assumed after a specific time in the prediction horizon and compares it with cases in what the load current is assumed known throughout the entire prediction horizon.

Abstract: Systems procured as digital twins of water networks are widely promoted for municipal use, yet evidence on whether they close the chain from sensor to decision in small utilities is scarce. We report a comparative operational audit of three publicly funded Internet of Things (IoT) deployments in Greek water utilities: Argos–Mycenae, Aigialeia and Souli. Using a harmonised twelve-month window and an explicit 0–4 rubric, we scored five sequential layers—sensing, transmission, data management, modelling and decision integration—and recomputed the utilities’ regulatory loss indicators. Modelling was never chain-limiting, but only one case evidenced a calibration. Data management was chain-limiting in all three cases and transmission in two: the Souli archive holds 240 one-minute records, 53 of 119 channels do not vary, and the tags needed to interpret that are absent; the Aigialeia daily series reaches 96.9% completeness but only 48 h are sub-hourly. Documentary decision traceability was level 1 throughout, bounded by the retrieval scope. The outcome holds under four alternative windows, and the statutory reporting artefacts disagree by 3.6% on real losses. The constraint is data stewardship rather than instrumentation, so the layer test transfers to other instrumented urban services.

Abstract: This paper is concerned with robust performance analysis for continuous time invariant systems governed by linear stochastic differential equations driven by statistically uncertain Ito processes. The uncertainty is understood as the deviation of imprecisely known probability distributions of the input disturbance from those of isotropic white-noise disturbances which, up to scaling, are organized as a standard Wiener process. Using Tustin's transform with a one-parameter family of conformal maps of the unit disk in the complex plane onto the right half-plane for discrete and continuous time transfer functions, the deviation from the nominal isotropic white-noise model is quantified by the mean anisotropy for the input of a discrete-time counterpart of the original system. The parameter of this conformal correspondence specifies the time scale for filtered versions of the input and output of the system, in terms of which the worst-case root mean square gain is formulated subject to an upper constraint on the mean anisotropy. The resulting two-parameter counterpart of the anisotropy-constrained norm of the system for the continuous time case is computed in state space by using the methods of the discrete-time anisotropy-based theory of stochastic robust filtering and control, originated by the author in the mid-1990s. These results are illustrated by an application to an electric circuit with inductive coupling and statistically uncertain noise.

Abstract: To address the drawbacks of horizontal- and vertical-axis wind turbines, a new concept has recently emerged. A novel active-axis wind turbine (AAWT) aims to reduce mechanical loads and structural costs when compared to conventional horizontal-axis wind turbines. The AAWT uses a pitch-tilt angle mechanism to balance aerodynamic lift and centrifugal forces acting on the rotor, thereby providing stable rotor rotation. Previous research on a small laboratory prototype AAWT established the relationship between pitch and tilt angles for quasi-static equilibrium. However, the transient dynamic behaviour about the tilt axis during movement between these states has not yet been investigated. This work develops a transient dynamic model of the AAWT pitch-tilt mechanism and controls the transition between the equilibrium states of aerodynamic lift and centrifugal moments. The cross-arm tilt axis is introduced as the first degree of freedom of the rotation system, triggered by a change in balance between the aerodynamic lift moment and the centrifugal moment about the tilt axis. A controller that combines feedforward and PD feedback is used to control the tilt transition and compensate for the lack of a mechanical damper on the tilt arm axis. To achieve a stable response across different rotor speeds, gain scheduling is applied to the controller’s derivative term. The results show that the PD controller can suppress oscillations in the transient response and provide smooth convergence to the new target equilibrium point. At different tested rotor rotation speeds, the settling time decreases with increasing rotation speed. The research provides a basis for future control-strategy experiments to achieve dynamic changes in tilt angle in the AAWT.

Abstract: As the penetration of distributed generation (DG) in distribution networks continues to grow, fault currents shift from a conventional unidirectional flow to bidirectional flow, and the accuracy and adaptability of existing fault location methods are severely challenged when feeder terminal units (FTUs) are subject to false alarms and missed detections. To cope with high DG penetration and FTU information uncertainty, this paper proposes a topology-constrained adaptive binary Archimedes optimization algorithm (TCAB-AOA). Fault location is modeled as a discrete combinatorial optimization problem: an expected fault current function accounting for current direction and information uncertainty is constructed; a hybrid hard-intersection–soft-voting strategy exploits network topology and FTU measurements to compress the search space; differential evolution (DE) and Lévy flight are fused as dual perturbation mechanisms to improve convergence stability; and a greedy cleanup step eliminates residual spurious fault sections. Simulations on the IEEE 33-node system under seven DG-integrated fault scenarios show that TCAB-AOA attains an average location accuracy of 90.0% (47.1 percentage points above the standard Archimedes optimization algorithm) and reduces the average number of misjudged sections to 0.20; in the multi-fault false-alarm scenario, it ranks ahead of particle swarm optimization (PSO), genetic algorithm (GA), and DE.

Abstract: Robust admissibility shows that the closed-loop poles remain within the prescribed region, but it does not determine whether small variations caused by tolerances and changes in operating conditions can lead to abrupt pole movement and deterioration of the dynamic behavior. This paper proposes a sensitivity-mapping method that complements root-contour analysis with modulus, phase, and total pole sensitivity. The local pole displacement is decomposed into radial and tangential components, while the largest singular value of a normalized pole-sensitivity matrix evaluates the worst-case combined influence of the uncertain parameters. The indicators are mapped onto the corresponding pole locations and are complemented by sensitivity measures and envelopes of the transient responses. The method is applied to an uncertain second-order control plant with a PI controller and seven tunings numerically confirmed as robustly admissible. The results reveal differences among the robustly admissible tunings, both in the local pole motion and in the time domain. The approach supports the selection among robustly admissible tunings by localizing fragility, determining the nature of the pole motion, and assessing its manifestation in the time-domain behavior.

Abstract: Residential microgrids are expected to enhance both environmental sustainability and disaster resilience in future housing developments. This study investigates two distributed control schemes for residential DC microgrids: an autonomous distributed cooperative control based on the state of charge (SoC) of battery-driven DC/DC converters, and an autonomous decentralized control based solely on feeder voltage, enabled by directly connected distributed batteries. A detailed DC microgrid model consisting of five houses was constructed in MATLAB/Simulink, incorporating PV panels with different orientations, small wind turbines, household load profiles, EV charging, and realistic meteorological data. The cooperative control scheme effectively stabilized feeder voltage, suppressed surplus renewable generation, and reduced reverse power flow to the utility grid. The decentralized scheme also maintained stable feeder voltage without device-to-device communication, although it resulted in greater reverse power flow due to less effective generation curtailment. In both schemes, houses near the grid-interconnection point experienced more frequent battery cycling, indicating the need for larger battery capacities at those locations. Simulations showed that installing approximately 20 kWh of battery capacity per house enables more than 80% of annual consumption can be obtained from renewable energy and allows continued operation during off-grid conditions with reduced consumption. These results demonstrate the practical feasibility of distributed-control residential microgrids.

Abstract: Ultrasonic welding of Carbon Fibre Reinforced Polymers (CFRPs) is a joining process that promises a high velocity and advanced welds. Furthermore, the process is suitable for large-scale automation projects that can accommodate the corresponding high-rate capability requirements. Continuous ultrasonic welding, in particular, is subject to specific boundary conditions that are constantly changing, owing to the progressing weld seam. For this reason, it is particularly important for applications in the aerospace sector that the continuous ultrasonic welding process is actively controlled, especially when a thorough and robust weld seam quality is required. This short communication introduces a novel control method where material feedback (temperature) is incorporated as a reference variable by using an infrared thermography camera as a feedback sensor in a closed loop controller.

Abstract: During the deadly hantavirus outbreak of May 2026, the loudest online response called it a hoax, while measles, the one rising threat, drew little attention. That inversion is the half of the infodemic we rarely measure: not amplified fear, but the quiet downplaying of a real risk. Detecting it requires what the field lacks: an external benchmark for the true risk. Anchoring one in the epidemiological and mass-gathering literature, with the 2026 FIFA World Cup run-up as a dated window, two coders judged Reddit claims about measles (high), hantavirus (low), and Ebola (near zero) for accuracy and direction (Cohen's kappa 0.72 and 0.68). Of 66 coded claims, 41 were inaccurate, and distortion ran both ways: 24 downplayed risk and 17 inflated it (exact binomial p = 0.35), with downplaying at least as common as inflating. Attention ran opposite to danger, hantavirus drawing the most claims (44 of 66) and measles the fewest (8): the infodemic did not mirror the epidemic. The curated, single-platform sample makes these patterns exploratory, but they suggest that defenses built against panic alone miss its mirror image, the denial of a real threat that prior work links to weaker protection among the most exposed.