The study aims to improve State of Charge (SoC) estimation accuracy and robustness under temperature fluctuations, measurement noise, and nonlinear battery behavior. This study proposes a hybrid intelligent framework for accurate SoC prediction in Lithium-ion batteries (LIBs) by integrating the Efficient Predefined-Time Adaptive Neural Network (EPTANN) with the Portia Spider Optimization Algorithm (PSOA). Because of its ability to simulate intricate nonlinear and temperature-dependent battery dynamics with fast, reliable predefined-time convergence, EPTANN is used. PSOA optimizes the network parameters by improving global search capability, enhancing convergence performance, and avoiding local minima. The combination of both methods enables more accurate, stable, and efficient SoC estimation compared to standalone approaches. The proposed method is evaluated using the Panasonic 18650PF battery dataset under different driving cycles (US06, UDDS, LA92) and temperature conditions (0°C, 10°C, and 25°C). Simulation results show that the proposed method outperforms conventional techniques such as SA-SVR-MEE-EKF, ABC-MLP, and CNN in terms of accuracy, error reduction, and robustness. An ablation study further confirms the effectiveness of the hybrid design by demonstrating the individual and combined contributions of each component. Overall, the proposed hybrid framework offers a dependable and computationally effective method for SoC estimation in sophisticated battery management systems. Its robustness across diverse operating conditions demonstrates its potential for practical LIB monitoring and energy management applications.
Bagasse ash is an agricultural waste material that can be used as reinforcement in metal matrix composites at low cost. The present study focuses on the development of Mg–Al₂O₃–bagasse hybrid composites by powder metallurgy and the determination of mechanical and microstructural properties of the fabricated composites varying the reinforcement content. The composites of Mg–Al₂O₃–bagasse of 90 wt.
High-frequency brake squeal is a persistent noise, vibration, and harshness concern in disc brake systems. This study aims to experimentally identify the root cause of brake squeal noise in a disc brake system exhibiting instability within the 3694–3719 Hz band, construct and validate a finite element model of the reference brake pad, and suppress squeal tendency through systematic geometric pad optimization, and experimentally verify the optimized design on the same test rig. A full factorial test campaign of 80 operating points (2–4 bar brake pressure, 50–200 RPM disc speed) is performed on a laboratory disc brake test rig. Experimental modal analysis under free-free boundary conditions identifies the root cause of squeal. A finite element model of the reference pad is validated using Latin Hypercube Sampling and Genetic Algorithm-based multi-parameter material calibration. Fifteen alternative pad geometries, generated through combinations of two chamfer types and three slot configurations, are evaluated via finite element modal analysis. A Squeal Suppression Performance Index (SSPI) formulated through the Weighted Sum Method with min–max normalization ranks the candidate designs against two competing criteria: critical frequency separation and friction surface area preservation. The top ranked design is manufactured by CNC machining, and its modal characteristics and squeal behaviour are measured under the same test conditions as the reference pad. Squeal is observed at 27 of 80 operating points as a narrowband tonal component invariant to operating conditions, identifying a 0.49
Slender structures subjected to wind or traffic excitation may exhibit nonlinear near-resonant responses governed by the proximity between excitation and natural frequencies. Within the lock-in region, frequency synchronization leads to statistically steady response with markedly non-Gaussian probability distributions. Just outside the lock-in boundary, additional spectral components emerge, the response becomes highly sensitive to small frequency detuning, and classical cyclostationary descriptions are no longer sufficient. The stochastic van der Pol aeroelastic model is analyzed in both single-degree-of-freedom and two-degree-of-freedom formulations. Stochastic averaging is used to derive reduced Fokker–Planck equations for the partial amplitudes in the stationary lock-in regime. Galerkin-based approximations are revisited and compared with finite element solutions of the Fokker–Planck equation and Monte Carlo simulations. Non-stationary near-boundary response is examined using time-dependent Fokker–Planck solutions and direct Monte Carlo simulations. For the stationary SDOF case, the analytical resonant solution, Galerkin approximation, FEM solution, and Monte Carlo estimates provide qualitatively consistent probability density descriptions, while quantitative indicators reveal the robustness of FEM and the limitations of polynomial Galerkin expansions away from exact resonance. In particular, the Galerkin approximation may suffer from conditioning difficulties and locally negative probability densities. Beyond the lock-in regime, both Fokker–Planck and Monte Carlo analyses reveal the transition toward quasi-periodic modulation and beating. The proposed framework provides a unified comparison of probabilistic solution techniques for near-resonant stochastic aeroelastic systems in stationary and non-stationary regimes. The results support the use of FEM as a robust deterministic reference for low-dimensional Fokker–Planck problems and clarify the practical limitations of Galerkin and trajectory-based Monte Carlo PDF reconstruction. The approach contributes to improved understanding of nonlinear stochastic vibrations near resonance and is relevant to wind- and traffic-induced vibration problems and aeroelastic applications.
The accuracy of the results in finite element method (FEM) simulations has always been limited by computational capacity. Improvements in computer performance have been offset by much larger FE models and more complex analyses. Rapid-iteration through simplified FE models allow engineers to make quick assessments of a system’s structural behaviour during the early stages of a project. This work proposes a FEM simplification method based on the dynamic behaviour of structural components. The simplification aims to evaluate the dynamic response of a satellite tray as a function of its type of interface with the surrounding structure. A threshold value is defined at which the interface can be considered continuous along its boundary. The Modal Assurance Criterion (MAC), commonly used to compare modal shapes between FEM models, is used to assess the equivalence between simplified and comparison models. A new indicator, the MAC Asymmetry Index (MACAI), is also introduced as a complementary indicator to characterize asymmetry patterns in modal correlation matrices and support the interpretation of modal comparison results. The study focuses on the CubeSat standard, analysing different tray aspect ratios and interface configurations. The methodology allows defining threshold criteria to ensure the validity of simplified models, enabling faster design iterations without compromising dynamic accuracy.
This fast transition to EVs drives up electricity demand and poses problems for the efficient management of EVs connected to renewable energy sources. The main objective of this study is to create an intelligent energy management framework, called Lyrebird Optimization Algorithm-Finite Basis Physics-Informed Neural Networks (LOA-FBPINNs), for optimizing energy efficiency, operating cost, and minimizing environmental emissions in EV charging stations (ECS) with battery energy storage systems (BESS). The research question is whether the proposed LOA-FBPINNs framework can deliver a superior control and energy management of the inverter compared to traditional optimization methods. The proposed framework is a hybrid of FBPINNs and the LOA. The optimal inverter control parameters are predicted using FBPINNs, which take into account the physical relationships governing the energy management system, and the optimal control signals are optimized using LOA. The framework is programmed in MATLAB and tested with Crow Search Algorithm (CSA), Honey Badder Algorithm (HBA), and Salp Swarm Algorithm (SSA). The proposed LOA-FBPINNs improves the energy efficiency to 96
This study numerically investigates the quasi-static bending, buckling, and free vibration behaviors of multi-morphology functionally graded porous (FGP) beams based on triply periodic minimal surface (TPMS) architectures, specifically focusing on Gyroid, Diamond, Primitive, and IWP topologies. The governing equations are established within the Euler–Bernoulli beam framework, explicitly incorporating the material gradation-induced neutral axis shift. The nonlinear distribution of porous density along the thickness is modeled, and three-dimensional finite element method (3D FEM) simulations are conducted alongside comprehensive parametric analyses to validate the theoretical formulations. Theoretical predictions show close agreement with 3D FEM results. The nonlinear distribution of effective axial stress under bending is strongly dependent on the gradient index. Furthermore, the neutral axis shift and boundary conditions significantly influence both the static and dynamic responses of the FGP beams. The natural frequencies, bending performance, and buckling loads of multi-morphology TPMS-based FGP beams can be effectively tailored by tuning the gradient index, morphology type, or base materials, demonstrating high potential for application-oriented structural design.
The wind turbine tower serves as the primary supporting structure of a wind turbine, and its stability directly determines the operational safety and service life of the unit. During operation, the tower is subjected to long-term complex loads such as wind loads and gravity, which can easily lead to loosening of the flange connection bolts. This may result in increased tower vibration or even collapse. However, existing vibration-based methods are often sensitive to environmental noise, while conventional graph neural networks may introduce redundant connections and inadequately represent relationships among modal features. To address this issue, this paper proposes a method for identifying loosening of tower flange bolts based on vibration signals, utilizing Multi-level Singular Value Decomposition (MSVD) for feature extraction and a Dual-enhanced Graph Neural Network (De-GNN) model. MSVD uses the singular value curvature spectrum to adaptively decompose vibration signals into modal sub-signals, enhancing damage-sensitive information and suppressing noise. The De-GNN model introduces the correlation dimension as a global information to enhance node features and adopts full connections within modes and corresponding-feature connections across modes to construct a physically meaningful graph topology. Experimental results on a 1:40 scaled wind turbine tower show that the proposed method achieves identification accuracies of 98.12
This study aims to develop a unified hygrothermoelastic model for thick circular plates subjected to internal heat sources, integrating dual phase-lag conduction, nonlocal elasticity, and memory-dependent derivatives to provide a basis for structural reliability assessments under coupled environments. The governing equations were formulated in cylindrical coordinates, and Goodier’s thermoelastic potential along with Michell’s function was decomposed into particular and homogeneous components. A sequence of integral transforms—specifically Laplace, Fourier sine, and Hankel transforms—was applied to reduce the coupled field equations to algebraic form. Homogeneous solutions expressed through Bessel functions and eigenfunction expansions were utilized to satisfy radial and axial boundary conditions, with orthogonality relations applied to determine the precise coefficients. The model demonstrated that moisture–heat coupling significantly modifies radial stress development within the thick circular plates. Furthermore, the analysis revealed that neglecting nonlocal elasticity or memory-dependent effects leads to an underestimation of peak stresses and misrepresents transient wave fronts during transport phenomena. The proposed formulation successfully captures finite-speed transport phenomena, scale-dependent softening, and hereditary relaxation effects, providing a reliable predictive model for composite structures used in aerospace rotors, automotive brake plates, and nuclear reactor components. Furthermore, the model provides a foundation for advanced structural design analysis for industrial and aerospace applications, enabling more accurate life-prediction and optimization of high-performance components under extreme thermo-mechanical loading.
Cross-condition remaining useful life (RUL) prediction of rolling bearings faces two coupled challenges: reliable initial degradation point (IDP) identification under strong noise interference, and negative transfer induced by static domain alignment strategies. Conventional methods ignore degradation stage priors and adopt fixed distribution matching, which readily distorts target degradation trajectories and deteriorates prediction performance when wear patterns diverge across operating conditions. This paper proposes an IDP-guided dynamic domain-adaptive prediction method. First, a robust health indicator is constructed via a sparse time-frequency autoencoder generative adversarial network, and a progressive IDP identification scheme integrating dynamic thresholding and curvature localization is developed to divide the bearing life cycle. Second, a RUL prediction network integrating enhanced residual separable convolution, convolutional block attention, maximum mean discrepancy and domain-adversarial learning is established. A dynamic alignment coefficient and decoupled collaborative prediction head are designed to suppress negative transfer. Experiments on PHM2012 and XJUT-SY datasets under both intra-condition and cross-condition tasks show that the proposed method outperforms representative comparison methods in prediction accuracy and stability. Compared with the CNN baseline, average RMSE and MAE are reduced by 42.4
Existing cold-rolling vibration models commonly treat the roll stack and moving strip as separate subsystems, limiting consistent assessment of cross-directional roll motion and transverse strip dynamics. This study develops an integrated analysis framework for a four-high cold rolling mill. Energy equivalence was used to derive lumped masses and stiffnesses for the segmented roll-stack and housing components, and an eight-degree-of-freedom vertical–horizontal roll-stack model was established. The strip was modeled separately as an axially moving, tensioned continuous beam with localized elastic support at the roll gap. Roll-stack stability was evaluated through a first-order generalized eigenvalue problem and the spectral abscissa. Galerkin discretization was used to determine strip frequencies and mode shapes. Operational vibration measurements provided frequency-domain comparison. Calculated frequencies of 131.8, 259.0, and 348.9Hz were compared with a measured component at 118.3Hz and band-center frequencies of 244.2 and 342.5Hz, giving relative differences of 11.4
This study presents a parametric investigation of the effects of suspension stiffness coefficient, vehicle velocity, and sprung mass on the vibration damping performance and energy consumption of semi-active and active suspension systems under three different road profiles. Both suspension systems are modelled within a quarter-car framework and controlled via PID-based strategies. Initially, the vibration damping performance of semi-active and active suspension systems is evaluated for nominal conditions under three different road profiles, and the results are compared with those of a passive suspension. Subsequently, a parametric analysis involving 343 combinations is conducted to systematically examine the performance and energy consumption of both suspension systems. The parametric analysis indicates that increasing suspension stiffness reduces vibration-damping performance and increases energy consumption in both controlled suspension systems. Vehicle speed also affects vibration attenuation and energy consumption, though its influence is weaker than that of stiffness and becomes non-uniform under random road excitation. The effect of sprung mass, meanwhile, depends on the road excitation characteristics. Overall, the results provide a systematic energy-performance perspective for evaluating and tuning active and semi-active suspension systems. For the active suspension, increasing the passive damping coefficient improves vibration-damping performance under road profiles, while its effect on energy consumption generally depends on the road profile. The semi-active suspension provides a more favorable compromise between sprung-mass vibration attenuation and electrical energy demand, particularly under repetitive and random road excitations. The active suspension offers bidirectional direct-force actuation and more confined wheel and suspension-deflection responses, but at a substantially higher energetic cost.
Microscale water films and mist on precision optical surfaces—such as camera lenses and imaging devices—can severely degrade image quality and sensing reliability, while conventional thermal, chemical and mechanical dehumidification methods may introduce thermal stress, modify surface properties, or cause abrasion. This study proposes a non-contact, atomization-based water removal approach that exploits the coupling dynamics between a piezoelectric ceramic and a glass substrate. The objectives are to establish a coupled vibration model of the bonded assembly, derive the threshold condition for droplet breakup induced by substrate vibration, and identify the excitation voltage and frequency that maximize substrate vibration and water removal efficiency. A two-degree-of-freedom lumped-parameter vibration model was established to characterize the electromechanical coupling between the piezoelectric ceramic and the 0.6 mm-thick circular glass substrate bonded with ultrasonic adhesive, and the condition for inertia-dominated droplet breakup (ω ^2X_0>6γ /ρ D^2) was derived theoretically. Three cylindrical piezoelectric ceramics of different dimensions were selected as actuators. An experimental platform incorporating a multifunctional main control board, an optical displacement sensor and an oscilloscope was built to measure the substrate vibration amplitude under varying excitation voltage and frequency. Water removal performance was quantified gravimetrically: a single 0.5 g water droplet was deposited on the substrate, subjected to 10 s of ultrasonic vibration, and each condition was repeated five times under controlled ambient temperature and humidity. The substrate vibration amplitude increased with excitation voltage and reached its maximum when the excitation frequency coincided with the first-order resonance frequency of the coupled system; a frequency detuning of only 1
In this paper, an Internet of Things (IoT) framework for early fault detection and predictive maintenance of induction motor is proposed based on Interpretable Generalized Additive Neural Network (IGANN). The research objective is to improve the accuracy of fault detection and overcome the problems of noisy sensor data, and optimal model parameter selection for reliable induction motor condition monitoring. The proposed IoT-MDS-EDFIM-IGANN framework is based on bearing vibration data as input. First, Constrained Normalised Subband Adaptive Filter (CNSAF) is used to remove noise and improve the quality of signal. Label Correlation Guided Borderline Oversampling (LCGBO) is then used to balance the dataset by creating synthetic samples close to decision boundaries. A Signed Cumulative Distribution Transform (SCDT) is used for statistical features extraction. The bearing faults are categorized as Inner Race Failure, Outer Race Failure, Rolling Element Failure, and Normal conditions by applying IGANN whose parameters are optimized by the Starfish Optimization Algorithm (SFOA). The model is implemented and tested using Python. The proposed method achieves an accuracy of 27.26
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that poses considerable challenges to healthcare systems worldwide. Early detection is critical for effective intervention; however current diagnostic procedures frequently lack the required precision and sensitivity. This research proposes a novel Hybrid Spatial Vision Boost classifier (HSVBC) model to enhance the accuracy and reliability of AD detection. The approach preprocesses MRI images using an Isotropic Adaptive Preserving Filter (IAPF), efficiently reducing noise while maintaining critical edges and structural information. The Dual Intensity Spatial Matrix (DISM) is proposed to extracts features by capturing complicated spatial patterns and intensity values, resulting in a comprehensive data representation. To solve the problem of high dimensionality, Hybrid Discriminant Analysis (HDA) is proposed to reduce dimensionality while preserving discriminative information, resulting in improved precision and reduced computational complexity. The XGBoost classifier is used for the final classification, improving AD prediction accuracy and efficiency. The proposed approach is evaluated using the Alzheimer Disease Neuroimaging Initiative (ADNI) dataset, which obtained measures such as accuracy 99.85
This paper investigates the dynamic characteristics of variable-section wing-like panels, with emphasis on vibration behavior and design-oriented performance regulation. Compared with traditional constant-section panels, variable-section configurations enable graded distributions of mass and stiffness, thereby improving material utilization, lightweight potential, and fatigue resistance. Governing equations for a cantilevered variable-section panel are derived based on classical laminated plate theory and von Karman nonlinear kinematics. The equations are discretized using Hamilton's principle in conjunction with the Rayleigh-Ritz method. Numerical simulations and experimental tests are conducted to validate the theoretical model and to examine the effects of boundary constraints and geometric parameters on vibration characteristics. The results demonstrate that boundary constraints predominantly govern the effective structural stiffness, while key geometric parameters such as length and thickness provide complementary control over the mass-stiffness distribution. Specifically, increasing panel length leads to a significant decrease in natural frequencies across all modes, whereas increasing panel thickness yields a slight upward shift in frequencies. The first four modes exhibit consistent trends in response to geometric variations. The proposed framework offers a practical pathway for targeted frequency tailoring and resonance avoidance. The findings provide theoretical support for lightweight optimization and vibration control of variable-section aerospace structures.
Accurately identifying learning disabilities, particularly dyslexia, remains challenging due to the limited effectiveness of existing computational methods in capturing subtle linguistic and cognitive variations in student responses. This study proposes an interpretable deep learning framework to improve the accuracy and reliability of learning disability classification by integrating advanced feature selection and generative modeling techniques. The proposed framework applies text preprocessing, including normalization, tokenization, and stop-word removal, followed by Cauchy Mutation-based Black Widow Optimization (CM-BWO) for selecting the most discriminative features. A Multi-Head Attention-based Progressive Generative Adversarial Network (MHAPGAN) is employed to generate realistic synthetic linguistic representations that capture underlying writing patterns associated with dyslexia. The generated features are used to train the classification model while enhancing interpretability by identifying influential linguistic attributes. The proposed framework achieved an accuracy of 98.9
Multi-axle air-suspension vehicles are subject to coupled pitch and roll motions caused by asymmetric road disturbances, steering maneuvers, suspension interactions, and modeluncertainties. Conventional centralized or fixed-topology cooperative control methods may not adequately address the time-varying interactions among suspension units or provide sufficient robustnessunder severe operating conditions.Multi-axle air-suspension vehicles are subject to coupled pitch and roll motions caused by asymmetric road disturbances, steering maneuvers, suspension interactions, and modeluncertainties. Conventional centralized or fixed-topology cooperative control methods may not adequately address the time-varying interactions among suspension units or provide sufficient robustnessunder severe operating conditions. This study aims to develop a distributed leader–follower cooperative control strategy for body-attitude stabilization of multi-axle air-suspension vehicles. The proposed method is designed to suppress vehicle pitch and roll oscillations, improve coordination among suspension units, and maintain robust performance in the presence of disturbances, uncertainties, and actuator constraints. A hierarchical control architecture is proposed by integrating a soft-assignment reference generator, an adaptive leader–follower communication topology, an extended state observer for disturbance and uncertainty estimation, and a Control Lyapunov Function-based Quadratic Program (CLF-QP) supervisory controller. A lower-level proportional–integral controller regulates the air-spring pressure. In addition, an adaptive edge-weight adjustment law is introduced to modify the interaction topology among suspension nodes. The stability of the closed-loop system is investigated theoretically in terms of uniform ultimate boundedness. The proposed strategy is evaluated through TruckSim/Simulink simulations and hardware-in-the-loop experiments under asymmetric bump disturbances and double-lane-change maneuvers. The proposed adaptive-topology multi-agent control strategy provides more effective suppression of body pitch and roll motions than the fixed-topology controller and the uncontrolled vehicle. The results demonstrate lower peak attitude deviations and reduced root-mean-square responses under asymmetric road excitation. During double-lane-change maneuvers at 70 and 90 km/h, the proposed controller improves roll-motion suppression and enhances coordination among the suspension units while preserving closed-loop stability. The control performance can also be adjusted through the controller parameters to achieve an appropriate balance between attitude regulation and control effort. The proposed hierarchical distributed control framework effectively improves the attitude stability of multi-axle air-suspension vehicles under road and steering disturbances. By combining adaptive communication topology, observer-based disturbance compensation, CLF-QP-based constraint handling, and lower-level pressure control, the method offers a robust and practically implementable solution for coordinated vehicle body-attitude control.
This study proposes a defect-assisted locally resonant phononic crystal (PnC) lattice to achieve simultaneous low-frequency vibration attenuation and piezoelectric energy harvesting by exploiting bandgap effects and defect-induced wave localization. Band-structure analysis combined with an out-of-plane polarization criterion is used to identify the flexural bandgap. A local defect is introduced by removing the two steel cylinders from the central unit cell, and a piezoelectric patch is placed near the defect region. The effects of defect position, external load resistance, and temperature-dependent material properties are numerically investigated. A surrogate-assisted multi-objective optimization framework integrating Latin hypercube sampling (LHS), gradient boosting regression tree (GBRT) modeling, and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) is further developed.Purpose A complete flexural bandgap of 502-817 Hz is obtained, and the introduced defect produces strong localized deformation and voltage amplification near the defect-mode resonance. The maximum output power reaches 42.3 μW at an optimized load resistance of 80 kΩ. The surrogate models achieve testing R2 values of 0.9954 for defect-mode frequency and 0.9762 for bandgap width, and representative Pareto solutions are verified by numerical re-evaluation. The proposed design provides an effective route for lightweight PnC-based piezoelectric energy harvesters with enhanced low-frequency energy localization and broadband vibration suppression, while the optimization framework enables coordinated tuning of defect-mode frequency and bandgap characteristics.
Using the measured modal information to update the initial finite element model, so that it can more accurately predict the dynamic characteristics of the actual structure, has always been an important issue of the structural dynamics model updating. Traditional methods fail to simultaneously correct multiple types of system matrices and tend to destroy the inherent symmetry and sparsity of structural matrices, which limits their practical engineering applicability. In this paper, by utilizing linear projection operators, an iterative method is constructed for updating damped gyroscopic systems. This method can not only update the mass, damping, gyroscopic and stiffness matrices simultaneously, but also incorporate the measured data into the finite element model. And during the updating processing, the symmetry and sparsity of the original structural matrices are preserved. The corrected results have intuitive physical significance and can be directly applied to the actual engineering problems. The effectiveness of the proposed method is demonstrated with three numerical examples.
