Abstract: The determination of the arc rating of arc resistant materials for the manufacture of personal protective clothing is conducted by measuring the incident and transmitted energies through calorimetry using thermocouples coupled to copper discs during the electric arc events. In this study, custom calorimeters were constructed incorporating both a thermocouple wire and an embedded optical fiber temperature sensor, and the arc ratings of different fabrics were determined in terms of arc thermal performance value (ATPV). The results revealed differences between measurements obtained with the two sensor types. Notably, the absence of electromagnetic interferences generated by the arc current and the enhanced time response achieved in the optical fiber temperature sensor signal led to an ATPV arc rating approximately 27% lower than that measured with the thermocouple. These findings underscore the importance of investigating the current methodology used for determine arc ratings, to ensure accurate measurement of incident and transmitted energy.

Abstract: Reinforcement Learning from Human Feedback (RLHF) has become the dominant post-training paradigm for aligning large language models (LLMs), yet it remains among the most energetically expensive workloads in modern AI infrastructure. Existing RLHF frameworks optimise primarily for throughput on homogeneous GPU clusters, neglecting the severe energy inefficiencies inherent in the multi-stage RLHF pipeline. We identify a fundamental and previously unexploited structural asymmetry: inference stages (Reward Model, Reference Policy, Critic) draw 60–75% less power per GPU than training stages, and their predictable single-pass computation maps naturally to FPGA accelerators. We present EcoRL-Sched, an energy-aware heterogeneous GPU–FPGA task scheduling framework comprising three tightly integrated innovations: (1) a power-profiling subsystem that characterises per-stage, per-model-size energy density via a novel Energy Density Index (EDI) metric; (2) an FPGA offloading engine on Xilinx Alveo U55C achieving 4.9× better tokens/Joule than H100 GPUs for reward and reference inference, running concurrently with GPU training via a latency-overlap protocol; and (3) an RL-based dynamic scheduler, a PPO-trained lightweight policy network, that uses real-time power telemetry and ROLL multi-task workloads to minimise pipeline bubbles and idle GPU cycles. Across 8B, 70B, and 405B parameter models on a 32-GPU H100 cluster, EcoRL-Sched achieves up to 14.6× throughput speedup, 38.4% energy reduction, 40.6% CO2 reduction, and 51% faster convergence on ROLL benchmarks, all without degrading model quality. Lifecycle analysis confirms net carbon benefits exceed FPGA manufacturing overhead by >30×.

Abstract: Deploying efficient sign language recognition models on edge devices advances inclusive, affordable, and privacy-preserving human–computer interaction. Yet most state-of-the-art architectures target server-class hardware and fail under the strict memory, computation, and energy constraints of microcontrollers. This work introduces S3D-Conv1D, a spatiotemporal architecture for isolated word-level sign language recognition, tailored for TinyML deployment. By factorizing spatial and temporal processing into lightweight two-dimensional (2D) convolutions followed by one-dimensional convolution (Conv1D) layers, the model eliminates recurrent dependencies and ensures deterministic, MCU-compatible computation. This design yields predictable latency, bounded activation memory, and stable model size, while supporting full post-training 8-bit integer (INT8) quantization and compatibility with TensorFlow Lite, CMSIS-NN, and NNoM. Three baselines, S3D, CNN+RNN, and attention-based embedded LSTM (e-LSTM), were evaluated using unified preprocessing, quantization, and profiling on WLASL100 and SemLex100 datasets. S3D-Conv1D achieved 98.4\% float32 accuracy with superior stability and generalization. After INT8 quantization, it retained accuracy within 0.18\% while compressing nearly 4× into sub-megabyte binaries (895.9 KB). Deployment profiling on a CPU-only runner, used as a proxy for microcontroller execution, showed S3D-Conv1D as the only architecture achieving full INT8 execution with real-time indicative performance (23.6 ms). These results demonstrate that efficient, edge-ready sign language recognition requires architectures designed around hardware constraints from the outset, rather than compressing high-capacity models.

Abstract: The ASF grid is one of the important methods for improving the performance of the eLoran system. When an ASF grid is established, users can calculate ASF through the ASF grid database and bilinear interpolation algorithms, thereby obtaining high-precision ASF values. However, establishing an ASF grid is a complex process, involving not only extensive data collection tasks but also the cumbersome processing of the collected data. This paper investigates and analyzes algorithms for establishing ASF grids under different conditions. When the number of test points exceeds the number of grid vertices, grids built using interpolation algorithms—particularly the inverse bilinear interpolation algorithm—perform the best, while those constructed using the Kriging algorithm perform second best, and the inverse distance interpolation algorithm yields the largest errors. When the number of test points is less than the number of grid vertices, grids constructed using the Kriging algorithm perform the best, while the inverse distance interpolation algorithm produces the largest errors. As a result, the inverse bilinear interpolation algorithm is the best choice, when the test points are more than the grid points. The Kriging algorithm is recommended, when the test points are sparse.

Abstract: Building spatial flow management is a critical link in ensuring the safe and efficient operation of spaces. Overcrowding in such spaces is likely to trigger a series of negative impacts, including delays and potential safety hazards. In this paper, we use the spatial flow estimation method to identify the future high-utilization space in the architectural design stage, so as to optimize the design scheme at the source and effectively alleviate the hidden danger of over-saturation of spatial flow in super high-rise buildings and transportation hubs. Firstly, the influencing factors of spatial flow are deeply analyzed in terms of the spatial structure and pedestrian demand based on fine microscopic simulation data of building pedestrians. Then, a space utilization intensity index is designed by introducing the visual integration degree of space syntax and the definition method of origin–destination (OD) influence scope considering obstacle derouting. Furthermore, the spatial flow is estimated based on the space utilization intensity index using regression analysis. Finally, taking the basement 1 floor of the Shenzhen Bay Super Headquarters Base C Tower connected with the metro as an example, the effectiveness of the proposed method is verified. The results show that the MAPE is 26%.

Abstract: This study develops a human-centered AI framework enabling rapid ecodesign prioritization for ESPR compliance while demonstrating Large Language Model (LLM) integration in sustainability strategy. Four-stage hybrid methodology combining LLM-assisted action identification (30 ESPR-aligned interventions) with Multi-Criteria Decision Analysis with Analytic Hierarchy Process (MCDA-AHP) is developed. Expert validation addressed LLM-driven interventions limitations with practitioners evaluating AI suggestions based on value chain context. The framework applied to two Italian BIPV SMEs demonstrated strategic differentiation based on Feasibility vs. Desirability vs. Affordability producing systematically different action portfolios within regulation-aligned aggregate structures. Sensitivity analysis showed 100% priority stability under ±10% AHP variations for priority 1, 3 and 4 actions and 82% for priority 2 actions, validating framework robustness. The framework's provide empirical evidence for augmentation-not-automation in AI-assisted strategic planning, contributing replicable methodology for responsible LLM integration across manufacturing sectors. Results demonstrate combining AI synthesis efficiency with human contextual judgment enables regulation-aligned, business-model-specific sustainability strategies.

Abstract: The integration of truck platooning and electrification presents a promising avenue for improving operational efficiency and environmental sustainability in freight transportation. Realizing the energy and cost saving as well as emission reduction benefits requires a holistic design of truck routing, scheduling, and platooning strategies that account for practical operational constraints. This study investigates the integrated planning problem of routing, scheduling, and platooning for a mixed fleet of conventional trucks (CTs) and electric trucks (ETs), referred to as mixed fleet truck platooning (MFTP) problem. The MFTP incorporates charging scheduling and key operational factors, such as platooning leader-follower positioning under the battery constraints of ETs and charging station availability and capacity, and the positional configuration of trucks within a platoon. The objective is to minimize the total operation cost of the MFTP system, including charging cost, fuel cost, travel labor cost, charging labor cost, and platoon formation labor cost, while ensuring timely arrivals across multiple origin–destination (OD) pairs. The proposed MFTP is formulated as a novel mixed-integer linear program (MILP). Extensive numerical experiments on the Illinois highway network are conducted to examine the effectiveness and efficiency of the proposed model. The findings shed light on planning mixed fleets of CTs and ETs with platooning, offering valuable managerial insights for decision-makers.

Abstract: Longwall paste backfilling mining is severely restricted in large-scale application due to its sequential mining-isolation-backfilling-curing operation mode, which leads to low production efficiency and poor economic feasibility. Taking the E1302-B paste backfilling face of Gaohe Coal Mine as the engineering background, this study systematically identified the key efficiency-restricting factors including low isolation efficiency, cumbersome backfilling process, prolonged paste curing time and insufficient system operation controllability, in view of the complex geological conditions of the face such as severe undulation of roof and floor and irregular cross-section. Technological innovations were carried out from four core dimensions: high-efficiency isolation, high-efficiency backfilling, accelerated curing and intelligent safety control. A self-adaptive mechanized isolation device was developed, a high-efficiency backfilling process with simplified isolation procedures was proposed, the optimal accelerator addition method with cross-blade mixing was determined, and an integrated intelligent monitoring system covering the whole process of equipment-pipeline-support was constructed. Eventually, a high-efficiency integrated technology system for longwall paste backfilling mining was formed. Industrial test validation demonstrated that the technical system significantly boosts the efficiency of isolation, backfilling and solidification in the backfill mining cycle, cutting the time of a single backfill mining operation cycle by 57%. The annual production capacity of the E1302-B face was increased to 0.81 Mt, with a comprehensive backfilling mining cost of 466.63 CNY/t, an annual economic benefit of 108.03 million CNY and a static investment return rate of 48.96%. The E1306 face achieved an even higher annual production capacity of 1.12 Mt with a static investment return rate of 74.94%. This technology system effectively breaks the efficiency and economic bottlenecks of traditional longwall paste backfilling mining, realizes the dual improvement of backfilling mining efficiency and economic benefits, and provides technical support and practical approaches for the large-scale application of longwall paste backfilling mining.

Abstract: Long-wavelength flank waviness plays a critical role in the excitation behavior of geared transmissions. While coordinate measuring machine (CMM) exports provide detailed geometric information, conventional evaluations typically focus on individual tooth curves and do not quantify circumferential inhomogeneity across teeth. This study introduces a tooth-to-tooth long-wavelength waviness inhomogeneity indicator (ΔW1) derived directly from Klingelnberg-style MKA plot files and demonstrates its behavior on a large industrial dataset comprising 3375 measured gear parts. Each flank curve was detrended using a second-order polynomial fit, and lobe-based waviness amplitudes (W1–W3) were extracted via sine–cosine projection. The proposed ΔW1 metric was defined as the difference between the maximum and minimum W1 values across measured teeth within the same part. To eliminate measurement edge effects, a mid-section evaluation (10–90% of the face width) was additionally performed. Population-level analysis revealed consistent separation between geometrically homogeneous and inhomogeneous parts, with ΔW1 values in the most critical components exceeding 7–9 µm after mid-section filtering. Unsupervised clustering based on ΔW1 and maximum W1 further distinguished a defect-prone subset of parts exhibiting systematic long-wavelength modulation patterns. The results demonstrate that circumferential waviness variability can be quantified using standard CMM outputs without additional hardware or specialized measurement procedures. The proposed indicator provides a practical geometric screening tool for large production batches and establishes a reproducible framework for linking detailed flank geometry to manufacturing consistency assessment. Although acoustic validation is outside the scope of the present work, the metric is intended as an NVH-relevant geometric risk indicator for future vibroacoustic correlation studies.

Abstract: The rheological behavior of chitosan–vanillin crosslinked olive oil–in–water emulsions (Φ=0.52) was systematically studied as a function of key processing (homogenization time and speed, reaction temperature) and compositional variables (chitosan concentration, vanillin-to-chitosan molar ratio, and Tween® surfactant) to optimize their performance as oleogel precursors. All emulsions displayed viscous-dominant behavior, with a characteristic inflection in the storage modulus slope at ~0.1 Hz, except for Tween®-containing systems, which superimposable flow curves confirmed non-thixotropic Herschel–Bulkley pseudoplastic behavior (n ≈ 0.73) was observed. Optimal homogenization conditions (4 min, ≥ 9,500 rpm) promoted microstructural refinement without compromising emulsion stability. Increasing reaction temperature to 55 °C, approaching the chitosan percolation threshold (~0.8–0.9% w/w), and a vanillin-to-chitosan molar ratio of 0.7 maximized yield stress (up to 14.21 Pa), consistency, and thermal robustness, attributed to enhanced Schiff-base crosslinking and network densification. Tween® 20 and Tween® 60 induced oscillatory stiffening but caused pronounced softening under rotational shear due to interfacial displacement effects, with Tween® 20 providing superior thermal stability. Overall, a surfactant-free formulation (0.9% w/w chitosan, molar ratio 0.7, 55 °C) yielded highly structured, gel-like emulsions, demonstrating enhanced suitability as templates for olive oil oleogel development compared to conventional stabilization strategies.

Abstract: Water scarcity has increased the need for efficient treatment technologies such as membrane distillation (MD). MD performance depends strongly on membrane fabrication parameters, particularly polymer concentration and nanoparticle incorporation, which control key transport and separation properties. This study considers fabrication of membranes using different concentrations of polyvinylidene fluoride (PVDF) with the incorporation of different types of nanoparticles to determine the optimum membrane formulation for membrane distillation applications. The results demonstrate that both PVDF concentration and nanoparticle type play a critical role in membrane performance in terms of permeate flux and salt rejection. Among the nanoparticles studied in this work, carbon nanotubes (CNTs) exhibited the most significant enhancement, leading to a substantial increase in water vapor flux while maintaining excellent separation efficiency. The optimized CNT incorporated membrane achieved approximately 99% salt rejection, with superior flux performance, indicating its strong potential for high-efficiency desalination and water treatment using membrane distillation.

Abstract: Lower-limb rehabilitation exoskeletons must balance biomechanical compatibility, structural safety, and low mass to enable practical, repeatable gait assistance. This paper proposes a planar pantograph-derived exoskeleton leg driven by a Chebyshev Lambda linkage and develops an integrated workflow from mechanism synthesis to manufac-turable optimization and experimental verification. A mannequin-coupled multibody model was built in MSC ADAMS to evaluate joint kinematics, end-point (foot) trajec-tories, and joint reaction forces under multiple scenarios (fixed-frame, ramp, stair as-cent, and inclined-plane walking). The extracted joint loads were transferred to a par-ametric finite element model in ANSYS Workbench, where response-surface surrogates and a multi-objective genetic algorithm (MOGA) were used to minimize mass under stiffness and strength constraints. For the optimized load-bearing link, the selected minimum-mass design reached a component mass of 0.542 kg while respecting the imposed structural limits, i.e., a maximum total deformation below 0.2 mm and a maximum equivalent (von Mises) stress below 55 MPa (e.g., ~0.188 mm deformation and ~39 MPa stress in the optimal candidate). A rapid prototype was manufactured by 3D printing and experimentally evaluated using CONTEMPLAS high-speed video tracking, providing measured X_M(t) and Y_M(t) trajectories and joint-angle histories for quantita-tive comparison with simulations via RMSE metrics.

Abstract: Virtual Power Plants (VPPs) face significant challenges in managing the uncertainty and variability of distributed energy resources (DERs), which can result in high trading risk and deter investment. This paper proposes and evaluates two advanced optimisation techniques—stochastic programming and robust optimisation—to derive risk-aware bidding strategies for VPP participation in the day-ahead and balancing electricity markets. These methods are benchmarked against a deterministic, expectation-based model. The novelty of this work lies in the comparative application of stochastic and robust frameworks to VPP bidding strategy design under real-world uncertainty, the introduction of scenario-based wind and conventional generation models, and the integration of energy storage into the optimisation framework to assess its impact on profitability and risk mitigation. Through a series of simulations using actual market data from the UK (Elexon), we evaluate three generation portfolio configurations—conventional, renewable, and aggregated. The results show that while stochastic optimisation consistently achieves the highest expected profit, the robust model ensures the highest minimum profit under worst-case conditions. Moreover, combining DER types and integrating battery storage further enhances profitability and reduces exposure to imbalance penalties. These findings provide valuable insights for the development of intelligent, risk-aware trading strategies for VPP operators.

Abstract:

Requirements for robustness and performance in the frequency domain in control theory are usually formulated as constraints on the modulus of complex functions describing the open-loop system, the sensitivity function, and the complementary sensitivity function. These constraints generate circular sets that can be interpreted as admissible or forbidden regions in the complex plane. In engineering practice, they are often treated as method-specific constructions, without clarifying the general geometric mechanism by which they arise. This study develops a geometric approach in which a broad class of frequency domain robustness constraints is represented as level sets of analytic and fractional-linear functions. The resulting circular sets in the Nyquist plane are characterized in a unified manner and transferred to admissible regions in the s-plane through preimage mappings. The approach is formulated entirely using complex transfer functions, without state-space representations, linear matrix inequalities, or optimization methods. Classical robustness measures, including gain margin, phase margin, and constraints on sensitivity and complementary sensitivity, are shown to be special cases of the same geometric structure. This interpretation establishes a direct link between frequency domain constraints and closed-loop pole locations, allowing a qualitative assessment of robustness and dynamic properties of control systems without introducing new stability criteria or design procedures.

Abstract: Experimental validation of dissolved oxygen (DO) control in aquaculture is often limited by biological variability, environmental factors, and pond hydrodynamics, which reduce reproducibility and hinder reliable assessment. To address this, we developed a laboratory-scale, control-oriented platform that minimizes external disturbances and enhances statistical reliability. Oxygen demand was emulated via chemical deoxygenation with sodium sulfite, so aeration experiments begin from near‑zero dissolved oxygen (DO). Sodium sulfite is added only during initialization; any residual persists briefly into the early closed-loop phase. Using this framework, On-Off control with hysteresis and discrete-time PID control were compared in terms of overshoot, rise time, settling time, and steady-state error. Under a confidence criterion, the PID controller required fewer repetitions than the On-Off strategy to achieve comparable reliability.

Abstract: This study proposes a surrogate-assisted evolutionary optimization framework for small dataset that integrates machine learning–based surrogate models with evolutionary algorithms for the aerodynamic optimization of a spiked blunt body in supersonic flow. A database of simulated cases covering a range of Mach numbers, spike length ratios (L/D), and diameter ratios (d/D) was used to train regression models and identify optimal geometries. Among the tested algorithms, the Gradient Boosting Regressor (GBR) achieved the best predictive performance (R² = 0.8909, RMSE = 0.00775), accurately capturing the nonlinear dependencies of the drag coefficient (C_d). Evolutionary optimization methods, including Differential Evolution (DE), Covariance Matrix Adaptation Evolution Strategy (CMA-ES), and Genetic Algorithm (GA), consistently converged to near-optimal configurations, with DE exhibiting the most stable behavior across Mach regimes. SHapley Additive exPlanations (SHAP) analysis revealed that (L/D) is the most influential parameter on C_d, followed by Mach number and (d/D), highlighting the dominant effect of geometric slenderness in drag reduction. The integration of data-driven modeling with evolutionary computation provides a robust framework for aerodynamic optimization, offering both predictive accuracy and physical interpretability. These results demonstrate the potential of hybrid Machine Learning-Evolutionary algorithms and CFD approaches to accelerate the design of high-performance configurations in supersonic applications.

Abstract: We develop a method for constructing Lyapunov functions via Semidefinite Programming (SDP) that certifies the stability of oscillatory systems with both Cartesian and angular variables. We utilize the theory of hybrid polynomials (also called mixed trigonometric-polynomials) introduced by Dumitrescu. We use this theory to convert Lyapunov and dual Lyapunov stability conditions for oscillatory systems into SDP problems. Solving these problems using standard convex programming solvers leads to expressions of Lyapunov densities and local Lyapunov functions for these systems, even without apriori knowing the invariant attracting set. To illustrate the applicability of our method, we consider the analysis of Kuramoto models and the state feedback design problem for an inverted pendulum on a cart. Specifically, we establish certificates of almost global synchronization (phase locking) for second-order Kuramoto models. The paper concludes by developing an SDP certificate that enables the design of a swing-up control for an inverted pendulum on a cart. For the analysis, we use our program vSOS-hybrid, based on CVX in MATLAB, openly available on GitHub.

Abstract: This study integrates in-situ Quantitative Infrared Thermography (QIRT) and Building Energy Simulation (BES) to optimize the energy performance of an existing multi-story residential building in a temperate climate. QIRT was utilized to diagnose thermal anomalies at the interfaces of uninsulated walls, RC skeleton and fenestration junctions, revealing significant thermal bridging and air infiltration while enabling the calculation of the Temperature Index (TI) at critical interfaces. A key finding of the non-destructive diagnostic phase was the discrepancy between in-situ (UINSITU) and theoretical (UCALC) thermal transmittance values, providing an empirical baseline for subsequent optimi-zations. A multi-objective analysis was conducted using genetic algorithms to evaluate 192 retrofit combinations, involving three insulation materials at four thicknesses and 16 glazing types. The impacts on primary energy consumption, CO₂ emissions, and 30-year global costs (per EN 15459-1:2017) were quantified under the volatile economic conditions. Findings indicate that the energy-optimal solution reduces primary energy by 53% and CO₂ emissions by 51%, while the cost-optimal configuration reduces global costs by 52% relative to the reference case. The Pareto analysis reveals a robust convergence between financial and energy efficiency targets, proving that deep retrofitting is an economically imperative strategy for achieving national decarbonization goals and the 2053 net-zero vision.

Abstract: This study investigates the mechanical response of bifacial glass-glass photovoltaic modules subjected to snow-type loading, with a particular focus on the influence of silicon cell spacing on global deformation and local stress distributions in the silicon layer. Five computational finite element models were developed which explicitly represent all laminate layers and discrete cell layout. The numerical results are interpreted within the framework of partial interaction and shear transfer between the glass plies, and are validated against previously obtained home conducted experimental observations. The results demonstrate that silicon cell layout has a pronounced effect on local tensile stresses in silicon cells and on the curvature distribution within the laminate, while its influence on the global kinematic response is less critical. The numerical analysis indicates that the relative displacements between the glass layers resulting from the flexibility of the adhesive bond play a critical role.

Abstract: Maleic anhydride (MAH) grafting is widely employed to compatibilise polylactic acid (PLA) in fibre-reinforced composites; however, the influence of reactant addition sequence during melt processing varies widely across the literature, with no clear consensus on an optimal approach. In this study, the effect of reactant addition sequence on the graft yield of MAH onto PLA was investigated using dicumyl peroxide (DCP) as an initiator. Four loading protocols were examined in which the order of addition of PLA, DCP, and MAH was varied using approaches commonly reported in the literature, while all other processing conditions were held constant. A strong dependence of grafting yield on addition sequence was observed, with values ranging from 0.12% to 0.51%, corresponding to more than a four-fold variation under otherwise identical processing conditions. Simultaneous addition of PLA, DCP, and MAH produced the highest grafting yield, attributed to a more effective utilisation of peroxide-derived radicals. These results demonstrate that reactant addition sequence is a critical processing variable governing MAH grafting efficiency.