Search Results (6,808)

The Transformer architecture drives modern Artificial Intelligence (AI), yet the physical principles that may constrain self-attention training remain poorly characterized. We develop a thermodynamic framework for attention training, drawing on the established Boltzmann correspondence between softmax attention and equilibrium statistical mechanics, and we propose a First Law analogue that decomposes the training energy budget into a heat term (the entropic cost of ordering attention) and a work term (the gain in mutual information about the target). From this framework we derive a Landauer-type bound on learning, which states that the loss reduction during training is bounded below by the entropic cost of structuring attention against thermal noise. The bound is satisfied across all configurations tested: 625 grid points spanning three datasets on a compact Vision Transformer trained from scratch (MNIST, CIFAR-10, and OrganAMNIST), and ten temperatures on a pretrained ViT-Small fine-tuned on Food-101. Reusing the same physical principles at inference time, we show that the thermodynamic work performed by each input patch provides a quantitative, energy-based measure of feature importance that outperforms standard attention weights and Integrated Gradients on ImageNet across pretrained ViT-Small, ViT-Base, and ViT-Large (22M to 304M parameters). The result is an integrated diagnostic framework that links phase structure, training-time bounds, and inference-time attribution within a single empirically falsifiable thermodynamic apparatus. Full article

This paper studies observer-based stabilization of a normalized incommensurate fractional-order discrete-time SI benchmark model for computer-virus propagation. The model is formulated with Caputo-like fractional-difference operators and allows the susceptible and infected compartments to have different memory orders. In contrast with a predictive malware-forecasting model, the proposed system is explicitly treated as a dimensionless benchmark for qualitative analysis and control design. To clarify how the benchmark can be connected to empirical cybersecurity data, the revised formulation includes a calibration and fractional-order selection procedure based on normalized infection telemetry, admissible parameter sets, and loss minimization. The incommensurate orders are therefore interpreted as identifiable modeling parameters, not as arbitrary constants. The plant, observer, and control laws are formulated on the integer update grid, and the memory terms are implemented through the equivalent Volterra-type convolution representation. A nonlinear Luenberger-type observer is proposed under infected-state measurements, which is justified as a detectability-based cyber-monitoring configuration rather than a full observability assumption. The observer gain design, the full-state feedback design, and the observer-based output-feedback design are derived from first-order linearized incommensurate fractional-order models. The resulting criteria are expressed through characteristic-root conditions associated with linear incommensurate Caputo-type fractional-order difference systems. The scope of the theoretical claims is made explicit: the results provide local linearized-design guarantees and do not establish global or semi-global nonlinear stabilization. The nonlinear residuals, measurement-noise channel, incomplete-measurement formulation, and limitations of the linearized characteristic-root approach are stated explicitly so that the numerical section can assess robustness, sensitivity, and the effective region of attraction of the nonlinear closed loop. Full article

This study proposes an enhanced physics-informed neural network (PINN) framework for predicting seepage fields around deeply buried tunnels with limited field measurements. Hard-constrained boundary enforcement via distance-function trial functions is introduced to exactly satisfy Dirichlet conditions on both the ground surface and tunnel perimeter, and Bayesian optimization automates loss weight tuning to replace costly manual calibration. A systematic evaluation of 15 sensor placement schemes demonstrates that the hydraulic head variance across monitoring points, governed by radial coverage distance, is the primary determinant of prediction accuracy—not the number of sensors or angular density. Remarkably, a strategically designed 12-point configuration outperforms 100 randomly distributed points under the idealized conditions studied, confirming that placement quality can dominate over quantity when physics-informed optimization is applied. Transfer learning experiments across 132 geometric configurations reveal a previously unreported geometric transition zone at D/R ≈ 13–15, where prediction errors exhibit a distinct non-monotonic peak. Finite element benchmarking confirms that this error peak stems from the learning characteristics of PINNs under competing boundary influences rather than from the physical complexity of the problem itself. High-density sampling effectively suppresses this peak error by 32% compared with sparse sampling. These findings establish quantitative sensor deployment guidelines for tunnel seepage monitoring and identify fundamental performance boundaries of physics-informed machine learning under geometry–physics coupling. Full article

This paper presents the process of gathering data for a flapping-wing micro air vehicle (FWMAV) using optical tracking and force sensors for subsequent dynamic modeling and simulation purposes. Tethered and clamped experiments were performed to track the vehicle’s overall motion, wing kinematic angles, and aerodynamic force patterns, while additional properties such as mass, inertia tensor, center-of-mass position, and short-period excitation frequency were also examined. The methodology includes the testing approaches, modeling choices, and error analyses applied to the measurements. The results demonstrate that both tethered and clamped configurations introduce key limitations, particularly for steady-state flight. Additional constraints include structural fragility (hindering high-frequency testing), over-simplified CAD geometry, and controller tuning issues on the tail. Based on the identified parameters and experimental datasets, a high-fidelity simulation model was developed in MATLAB to serve as a platform for future control and flight envelope studies. Overall, the combination of optical tracking and force sensing provides a structured framework for linking experimental data to physical models, laying the foundation for future improvements in ornithopter modeling and testing. Full article

Structural damage classification in composite UAV wings is a key challenge in Structural Health Monitoring (SHM), particularly under barely visible impact damage conditions. Fiber Bragg Grating (FBG) sensor networks provide high-resolution strain data; however, systematic experimental benchmarking of lightweight neural architectures trained on real FBG datasets remains limited, especially under sensor degradation scenarios. This work presents a four-phase benchmarking study of Multilayer Perceptron (MLP) configurations using strain measurements from a composite UAV wing instrumented with 32 FBG sensors across five damage states and 210 loading experiments. The framework evaluates optimization strategies, hyperparameter sensitivity, architectural depth, and robustness under controlled sensor dropout, Gaussian noise, and wavelength drift perturbations. Results indicate that compact architectures with progressive dimensional reduction (256–128–64) trained using adaptive optimizers (AdamW and Nadam) achieve the best balance between macro-F1 performance (up to 0.85 during validation), stability, and computational efficiency. Robustness analysis shows gradual performance degradation under sensor loss, suggesting distributed strain-field learning. These findings provide practical guidelines for selecting computationally efficient and robust neural models for deployable FBG-based SHM systems in aerospace applications. Full article

Vent holes in tire molds typically exhibit large depth-to-diameter ratios (25–50) and variable drilling angles, both of which increase the risk of drill-bit breakage during automated drilling. To address this problem, this study develops a robotic drilling system consisting of a 6-DOF industrial robot and a dedicated end effector integrating a spindle unit, a linear feed unit, and a telescopic drill-bushing unit. A GRU-based feed-torque model predictive control method (GFTMPC) is proposed for robotic small-diameter deep-hole drilling, which achieves flexible control by integrating angle-aware feed-torque modeling with constrained MPC-based feed-rate optimization. The resulting GRU-based feed-torque model (GFTM) is embedded in the MPC framework for torque prediction and achieves an R² value of 0.9682. Under identical simulation conditions, GFTMPC reduces the RMSE of the feed-rate increment by 34.82% and the saturation ratio of the feed-rate increment by 90.78% relative to a PID baseline, indicating smoother feed regulation and fewer abrupt control actions in simulation. Comparative engineering experiments further suggest that, under the tested robotic configurations, adaptive feed-rate regulation by GFTMPC is an important contributor to improved tool life and drilling reliability. Hole-diameter measurements show deviations ranging from +0.03 mm to +0.11 mm, which were considered acceptable for the subsequent work steps in this application. Engineering application results show that robotic drilling increases daily throughput per worker by 71.38% and the average number of holes drilled per bit by 237%. Full article

This study investigates steady-state heat transfer in a three-layer cylindrical system with angular non-uniformity of the temperature field. For the considered geometry, a mathematical model of heat conduction is formulated in cylindrical coordinates with piecewise constant thermophysical properties and continuity conditions at the interfaces between layers. The direct problem is solved analytically using a Fourier series expansion of the temperature field with respect to the angular coordinate. Based on experimental temperature measurements obtained for various configurations of soil layers, an inverse problem is formulated and solved to reconstruct the thermal conductivities of the individual layers and the heat transfer coefficient at the external boundary. To stabilize the solution, a regularized least-squares approach is employed. The convergence of the recovered parameters with respect to the harmonic number is analyzed, and the averaged reconstructed values are compared with the exact parameters used in the direct problem. The obtained results demonstrate the stability and accuracy of the proposed method, confirming its applicability to the identification of thermophysical parameters in multilayer soil systems based on experimental data. Full article

This article presents a comprehensive analytical, numerical, and experimental study of the amplification and detection of weak signals in magnetically coupled electromagnetic systems, using an architecture consisting of three magnetically coupled coils. A rigorous mathematical model of the system is developed, which includes the formulation of the mutual inductance matrix and a state-space representation that captures the dynamic interaction between the coils. It is important to note that the electromagnetic subsystem is linear and that the stochastic resonance effect is achieved by incorporating an external nonlinear bistable element. In this configuration, a weak periodic signal below a threshold is applied to the primary coil, while a controlled source of Gaussian white noise is injected into a secondary coil. A third coil functions as a sensing element, capturing the superimposed magnetic response resulting from coupling effects. The voltage induced in the sensor coil is subsequently processed by a bistable nonlinear element implemented via a Schmitt trigger, which provides the nonlinearity and bistability necessary to enable stochastic resonance and the detection of the weak periodic signal. The conditions of the SR are analyzed in terms of noise intensity, coupling coefficients, and system parameters, highlighting the existence of an optimal noise level that maximizes the signal-to-noise ratio (SNR) at the output. A detailed simulation framework has been developed in MATLAB/Simulink, enabling a systematic exploration of the parameter space and the validation of theoretical predictions. The simulation results are further supported by experimental measurements obtained from a physical prototype, which show agreement with the proposed model. The main contribution of this work lies in demonstrating that magnetically coupled electromagnetic structures can effectively interact with nonlinear bistable elements to exploit stochastic resonance in the detection of weak signals, even when the electromagnetic domain itself remains linear. The results demonstrate that magnetic coupling is an effective mechanism for mediating constructive interactions between noise and weak signals, thereby improving the detection of the latter. These results extend the applicability of stochastic resonance to hybrid electromagnetic systems and demonstrate its relevance in practical applications. Potential applications include ultra-sensitive magnetic detection, low-power signal detection, magnetic transducers, and robust signal recovery in noisy electromagnetic environments, particularly in contexts where conventional linear amplification fails. Full article

Polymeric membranes are widely investigated for CO₂ separation; however, their performance is often limited by the permeability–selectivity trade-off. Incorporating metal–organic frameworks (MOFs) and designing composite membrane architectures are promising strategies to overcome these limitations. This study aims to evaluate the effect of incorporating MOF-74 (Cu and Ni variants) into a polydimethylsiloxane (PDMS) selective layer supported on a polysulfone (PSF) membrane for enhanced CO₂/N₂ separation performance. Dual-layer PDMS/PSF composite membranes were fabricated via phase inversion for the PSF support, followed by solution casting of the PDMS/MOF layer. The developed membrane architecture introduces a synergistic design that combines the mechanical robustness of PSF with the selective transport capability of PDMS and the strong CO₂ affinity of MOF-74, offering an effective strategy for improving gas separation efficiency. Gas permeation performance was assessed using single-gas CO₂ and N₂ measurements at feed pressures of 2–5 bar. The incorporation of MOF-74 improved CO₂ transport properties, with the 1 wt.% Cu-MOF-74 composite membrane achieving a CO₂ permeance of 912.5 GPU and a CO₂/N₂ ideal selectivity of 94.75. The dual-layer configuration significantly enhanced permeance compared with unsupported mixed-matrix membranes while maintaining selectivity. Additionally, the composite membranes exhibited improved mechanical strength due to the PSF support layer. The findings demonstrate that dual-layer PDMS/PSF composite membranes incorporating MOF-74 provide a promising proof-of-concept approach for improving CO₂ separation performance. Further studies involving mixed-gas testing and long-term stability are required to assess their practical applicability. Full article

The piping systems are critical non-structural elements (NSEs) whose seismic performance directly affects the post-earthquake functionality of essential facilities. However, current seismic design provisions for such systems remain largely empirical, and behavioural factors are rarely calibrated using performance-based methods. This study implements an FEMA P695-inspired framework to calibrate the behaviour factor $\left(q_a\right)$for the installation of sway-braced suspended piping restraint systems in following the force-based requirements specified in Eurocode 8. The representative piping archetypes were developed and analysed using non-linear time history analyses under multiple seismic intensity levels derived from the floor response spectra (FRS) of prototype-reinforced concrete buildings. Fragility curves for two limit states were derived with displacement ductility adopted as the engineering demand parameter (EDP) and peak floor acceleration (PFA) used as the intensity measure (IM). The results show that increasing $\left(q_a\right)$ systematically shifts the fragility curves towards lower median PFA values, indicating higher seismic vulnerability at larger behaviour factor values. The effect of piping layout configuration was of secondary importance compared to the applied reduction factor. The implemented approach provides a rational basis for selecting behavior factors consistent with explicit performance objectives and supports further development of performance-oriented seismic design procedures for non-structural systems. The results show that increasing the behaviour factor $\left(q_a\right)$ leads to a systematic shift in the fragility curves towards lower median PFA values and a noticeable increase in the dispersion of the response. A quantitative analysis shows that increasing the behaviour factor $\left(q_a\right)$ from 1 to 4 results in a reduction of up to approximately 60% in median PFA, highlighting a significant increase in seismic vulnerability at higher behaviour factor values. Full article

In ice hockey, automatic activity detection using wearable sensors and machine learning could provide objective feedback to support coaches and players during performance evaluation. The primary objective was to assess the predictive ability of a deep learning model to recognize common ice hockey stick striking actions (passing, shooting) from inertial measurement unit sensors. This study implemented a fully connected convolutional neural network model to classify seven ice hockey-related technical actions (wrist shot, slap shot, backhand shot, one-timers, pass, other, and rest) using acceleration data via two setups: an all-sensor configuration (17 sensors) and a hands-only sensor configuration (2 sensors) in 43 elite players. Data were split into 80/20 train/test sets, with a five-fold cross-validation applied to the training data. The train/test split was repeated 10 times with different random splits to assess stability of results. The model achieved high classification accuracy, with the all-sensor model reaching an average F1 score of 95.0 ± 3.0% and the hands-only model achieving 93.5 ± 1.6%. These findings support the use of convolutional neural networks for automatic shooting action classification in ice hockey and highlight the feasibility of using minimal sensor configurations, such as sensor-integrated gloves, for real-world applications. This approach could further enhance training practices by providing objective performance metrics and allowing coaches to deliver data-driven feedback to players. Full article

Academic stress in higher education is strongly influenced by workload structure and scheduling decisions, yet academic planning sometimes remains static and does not incorporate behavioural or physiological indicators. While existing research focuses on stress measurement and prediction, these approaches are rarely integrated into decision-support mechanisms capable of restructuring academic schedules. This work introduces a Course Digital Twin (CDT) framework that integrates multimodal student data with simulation-based academic planning. The proposed system models course scheduling as a decision-support problem, where alternative configurations are evaluated using a structured stress model combining wearable-derived physiological signals, self-reported stress measures, and contextual academic workload indicators. The framework employs a hybrid approach in which machine learning is used for physiological stress estimation, while schedule adaptation is performed through transparent rule-based mechanisms. The system was implemented as an end-to-end platform including mobile sensing, course configuration interfaces, and instructor analytics dashboards, and was evaluated through a pilot deployment across multiple postgraduate courses. Preliminary results indicate that simulation-based schedule adjustments are associated with reductions in projected peak stress levels and improved workload distribution patterns. The findings demonstrate the feasibility of integrating multimodal stress modelling and Digital Twin simulation into academic planning workflows. The proposed framework provides a foundation for future stress-aware scheduling systems, although further large-scale validation is required to establish its effectiveness and generalizability. Full article

Cage-based tilapia farming is highly vulnerable to rapid variations in water-quality parameters, particularly dissolved oxygen (DO) fluctuations, which can cause fish stress, fish mortality, and economic losses. In this study, we developed and field-evaluated an Internet of Things (IoT)- and LoRa-based water-quality-monitoring and aeration-actuation system for open-water tilapia cage farming. The system consists of distributed control nodes, a main node, a cloud database, and a mobile application for real-time monitoring of DO, pH, and water temperature, as well as remote and automatic oxygen-pump actuation. An automatic probe-lifting mechanism is integrated into the control node to reduce probe-submersion duration and mitigate the risk of sensor fouling during field operation. Field validation showed that the node equipped with the probe-lifting mechanism achieved better agreement with the reference instruments than the continuously submerged node, particularly for DO measurement, with RMSE values of 0.186 mg/L and 0.683 mg/L, respectively. A communication-performance evaluation showed 100% packet reception up to 1640 m, whereas packet reception was reduced at the longest tested distance of 2290 m, indicating that the field-deployment range should be interpreted cautiously under the tested LoRa configuration. Detection-latency experiments showed sub-second responsiveness, with average delays of 208.6–289.7 ms for single-hop communication and 438.9–529.4 ms for two-hop communication. Expert evaluation and farmer satisfaction assessment indicated positive perceptions of the system’s usability and practical relevance. However, the study has several limitations, including the short field-validation period, limited sensor replication, and a lack of direct fish production outcome measurements, which should be considered when interpreting the findings. Overall, the proposed system provides a practical platform for water-quality monitoring and aeration actuation in cage-based tilapia farming. Full article

The suction pile well construction technique is increasingly adopted in deepwater drilling projects. The soil–structure interaction mechanism during the penetration and installation of the wellhead suction pile in clay is complex. Given the critical demand for precise installation outcomes in engineering practice, the influence of penetration velocity on installation performance requires significant consideration. Through scale-model experimental methods, various penetration velocities were configured primarily by adjusting suction pump flow rates. The influences of these velocities on penetration resistance, penetration depth, and related metrics were systematically assessed. A case study was conducted based on the engineering parameters of a wellsite in the South China Sea. A theoretical algorithm for WSP penetration resistance was developed and subsequently refined through experimental data. Coefficient optimization was established via theoretical assessment of strain-rate dependency and experimental data calibration. The optimized algorithm demonstrated strong agreement with field measurements, achieving a coefficient of determination ($R^2$) exceeding 0.9. Compared to conventional theoretical approaches, it incorporated explicit consideration of penetration velocity. The analysis indicates that in soft clay, the penetration resistance of wellhead suction piles exhibits significant sensitivity to penetration rate, increasing with higher velocities. The influence of penetration rate on penetration depth is relatively weak. This computational approach offers design guidance for installation procedures and enables the implementation of the suction pile well construction mode in the South China Sea. Full article

Geometric errors in CNC machine tools accumulate along the tool path and directly affect machining accuracy. Traditional serial-chain-based volumetric error models, such as those based on the homogeneous transformation matrix (HTM) or screw theory, often exhibit ambiguous geometric definitions, weak traceability to measurement points, and increased computational cost due to repeated coordinate transformations and inverse mappings, limiting their suitability for real-time control. To overcome these challenges, this study proposes a parallel-type unified error vector transfer (EVT) framework, based on the Abbe and Bryan principles. In this framework, axis error motions are directly expressed as vectors and transferred to the tool center point (TCP), where they are superimposed to obtain total error contributions. Building on this principle, a unified normal volumetric error model (NVEM) is formulated using survival and sign factors. The unified NVEM is applicable to various types of three-axis machining centers, including horizontal configurations. In other words, differences in coordinate system definitions can be reconciled through coordinate transformation, allowing the unified NVEM to be consistently applied. Furthermore, a real-time error compensation controller (RECC) is embedded into the CNC kernel to compute compensation values within each interpolation cycle, ensuring deterministic and low-latency operation without external computation. Experimental validations on an XYFZ-type vertical machining center demonstrate that the proposed framework improves positioning accuracy by more than 72% and machining accuracy by 60.4%. These results confirm the feasibility, efficiency, and universality of the parallel-type unified EVT framework for real-time volumetric error compensation. Here, ‘parallel-type’ denotes the parallel superposition of independent error vector contributions, rather than a parallel kinematic mechanism. Full article