The protection of underground infrastructure from modern missiles has become a critical engineering challenge for countries involved in military conflicts. This study presents a numerical investigation of the stress–strain response and blast resistance of underground tunnel structures subjected to a ballistic missile impact. The analysis was conducted for typical soil conditions of the Kyiv region (Ukraine), where sandy soils prevail at shallow depths. ANSYS Explicit Dynamics, combining Eulerian and Lagrangian solvers, was used. The propagation of blast waves in the soil mass, the formation of a blast crater, and the resulting stress distribution were analyzed for cases without protection and with reinforced concrete protective slabs of varying thicknesses. The results demonstrate that surface protection significantly alters blast wave characteristics and reduces peak stress in the near-surface zone, although excessive slab thickness may shift stress concentrations to a deeper zone. An express assessment method was proposed, based on the ratio between peak stresses in the soil and tensile stresses in the tunnel’s support, enabling rapid estimation of safe depth and required protective slab thicknesses. The proposed approach can contribute to rapid decision-making for the protection and design of underground infrastructure.
Concrete structures form the backbone of modern civil infrastructures, yet they are inherently susceptible to cracking over time due to mechanical stress, environmental exposure, and aging. Undetected cracks can escalate into catastrophic structural failures, making timely inspection critical. Traditional manual inspection methods are labor-intensive, subjective, costly, and unsuitable for large-scale deployment. Unlike prior surveys that treat vision-based crack detection in isolation, this paper positions automated crack detection as one component of a broader vibration-aware, multisensor structural health monitoring (SHM) ecosystem, in keeping with the scope of this journal. We present a critically synthesized review of automated concrete crack detection, spanning classical image processing and machine learning, convolutional neural networks, object detection, semantic segmentation, Vision Transformers (ViTs), generative adversarial networks, Explainable Artificial Intelligence (XAI), and, distinctively, vibration-vision sensor fusion, Digital Twin (DT) and physics-informed frameworks that connect surface cracking to internal structural state, emerging vision-language and foundation models (SAM, Grounding DINO, CLIP, and GPT-assisted inspection), and depth-aware 3D crack characterization via stereo vision, structured light, and LiDAR. Rather than only tabulating reported metrics, each methodological family is critically appraised along robustness, computational cost, practical applicability, and scalability, and every method-specific section closes with an explicit synthesis of strengths, weaknesses, and open failure modes. We further address two dimensions largely absent from earlier reviews: the severity of a crack is shown to be inseparable from its structural and environmental context (a small crack in a seismically active or dynamically loaded structure is not equivalent to a similar crack in a quiescent one) and the environmental operating condition (above-water, underwater, shadowed, or variably illuminated) fundamentally changes which detection strategy is viable. We have found that CNN-based transfer learning methods routinely exceed 99% classification accuracy on curated benchmarks, YOLO-based detectors provide real-time localization with competitive mAP scores, ViTs and hybrid architectures offer superior global feature modeling, and XAI tools such as Grad-CAM and SHAP enhance interpretability required for regulatory compliance, but all of these figures should be read against the robustness, cost, and generalization caveats we discuss explicitly. A dedicated comparison against existing review papers clarifies the specific gaps this survey closes, and quantitative bibliometric figures summarize publication trends, architectural evolution, dataset popularity, and accuracy-versus-year patterns across the literature. This survey concludes with a substantially expanded, author-articulated research agenda covering domain generalization, sensor fusion, DT-enabled predictive maintenance, and regulation-compliant explainability for next-generation automated structural inspection systems.
Barrel wear leads to a degradation in the ballistic performance of projectiles, conventionally manifested as a reduction in projectile muzzle velocity and a decrease in firing accuracy. Based on the actual wear distribution of the gun barrel, this paper establishes a parametric model of the worn barrel. By integrating the coupled effects of propellant combustion and mechanical interaction, a dynamic model of the barrel-projectile interaction across four life periods is developed. The underlying mechanisms between barrel wear, interior ballistic performance degradation, and projectile muzzle state are systematically investigated. The strong agreement between simulation and experimental results verifies the accuracy of the proposed parametric modeling method and the coupled model. The results indicate that in the later periods of barrel life, the projectile’s muzzle velocity does not exhibit a significant decline; however, notable changes occur in the projectile’s muzzle spin rate and disturbance. The decrease in the projectile’s muzzle spin rate and the increase in disturbance are identified as the primary reasons for barrel end-of-life.
This study presents a combined experimental and numerical investigation of the dynamic behaviour of a commercial prismatic lithium-ion battery module. The primary objective was to characterise the module’s modal properties and to quantify how experimental choices and modelling assumptions influence modal identification and numerical validation. A preliminary finite element (FE) model–guided sensor placement and excitation strategy and an extensive experimental modal analysis (EMA) campaign, using instrumented hammer impacts and a limited number of accelerometers, provided time-domain and power spectral density (PSD) measurements. To improve robustness, modal frequencies from multiple tests were combined using coherence-weighted averaging. Results show a high modal density between roughly 400 and 550 Hz: the first three experimentally identified modes occur at ≈399 Hz (torsion), ≈478 Hz and ≈550 Hz (bending). The FE model reproduced the main modal patterns but exhibited a swap between the second and third modes, likely due to geometric simplifications and mass/stiffness approximations. The work also reports practical best practices for modal analysis at the module level (suspension strategy, impact location, hammer tip selection and sparse-sensor layouts) and discusses PSD/time-domain analysis for the chosen mode shapes. Finally, the module model provides a suitable foundation for future scale-up studies. In particular, the use of reduced-order models and superelements is identified as a promising approach to extend the methodology to full battery packs while preserving accuracy and limiting computational cost. The combined EMA–FEM approach provides actionable guidance for EV designers to avoid dangerous resonance bands in battery systems.
It proposes a frequency-tuning method for ultrasonic horns based on variable mass, which involves attaching a mass ring (hereinafter referred to as the frequency-tuning ring) to the horn structure. By fixing different mass rings at the same position or adjusting the position of the mass ring, the equivalent mass and equivalent stiffness of the horn are regulated to achieve frequency tuning. Taking a three-section composite ultrasonic horn (its mass is 526 g) with resonant frequency of 20 kHz as an example, the influence of frequency-tuning rings with different masses fixed at various positions on the resonant frequency was calculated. The results show that after attaching a frequency-tuning ring with a mass of 120 g and a length of 20 mm, the frequency of the horn can vary in the range of 15.71–21.42 kHz. The calculated values are consistent with the test results. The study also found that the resonant frequency “rises instead of falling” when the ring is placed near the displacement node. The energy method was used to calculate the equivalent mass and equivalent stiffness of the horn, and this phenomenon was explained.
To address the limitations of traditional convolutional neural networks (CNNs) in extracting discriminative fault features under complex working conditions and the feature coupling problem in serial attention mechanisms, this paper proposes a rolling bearing fault diagnosis method based on the parallel dual-path attention multiscale ASPP network (PDPA-MSANet). The method constructs a parallel dual-path attention module, which independently models channel attention and spatial attention from the same feature map to enhance the representation of key fault features. In addition, a multiscale atrous spatial pyramid pooling (ASPP) structure is introduced. Convolutions with different atrous rates are employed to capture features across multiple receptive fields and improve the extraction of complex fault-related features. Experiments on the Case Western Reserve University (CWRU) bearing dataset show that PDPA-MSANet achieves an average diagnostic accuracy of 99.4%. Further tests on a self-developed rotating machinery fault experimental platform under variable-speed conditions demonstrate that the model maintains high diagnostic accuracy and good generalization ability. Cross-speed validation results further indicate that the proposed model can still achieve stable diagnostic performance when the training and testing samples are collected under different speed conditions. Ablation experiments further verify the contribution of the PDPA mechanism and ASPP structure to feature representation and diagnostic performance. These results suggest that the proposed method can effectively improve the fault diagnosis performance of rolling bearings under different working conditions, providing a feasible technical solution for intelligent operation and maintenance of rotating machinery.
In the factory inspection of RV reducers, traditional wired sensors face issues such as limited wiring, signal attenuation, and the lack of quantitative standards for manual judgment, which can easily lead to detection errors. This paper develops a wireless vibration acceleration sensor that can be directly magnetically attached to the output shaft end of the reducer. The RV-20E reducer is used as the subject of research for its factory performance inspection application. First, to address radial signal direction aliasing caused by sensor rotation with the shaft, a rotational-speed-adaptive synchronous demodulation method combined with a second-order Butterworth IIR filtering algorithm was proposed to decouple and separate horizontal and vertical radial vibration signals in the rotation coordinate system. Second, based on large-scale factory inspection data, we analyzed the time domain, frequency domain, and displacement signal characteristics under normal operation, central shaft runout, and misalignment conditions. Factory-level multi-indicator quantitative evaluation thresholds and graded alarm rules suitable for direct industrial implementation were established. Finally, dedicated host computer software was developed to realize real-time vibration visualization, intelligent graded alarming, and full-process data traceability. The study demonstrates that the wireless shaft-end direct measurement solution effectively avoids signal attenuation caused by the multilayer structural transmission path of fixed-end sensing schemes, resulting in a significant increase in vibration signal amplitude. The established quantitative standards effectively differentiate typical faults and prevent misjudgments caused by manual operations. This research provides standardized automated solutions for RV reducer factory inspection, ensuring measurement accuracy and consistency. The developed equipment features easy installation and strong operational adaptability, offering substantial engineering value in fault monitoring applications for industrial robot reducers and other gear transmission systems.
Deep learning has demonstrated remarkable potential for intelligent gearbox fault diagnosis; however, its effectiveness remains constrained by two critical challenges: the scarcity of labeled fault samples and the degradation of diagnostic accuracy under nonstationary operating conditions. Conventional supervised approaches rely heavily on abundant labeled data, which are often difficult or costly to obtain due to the rarity of severe fault occurrences. Moreover, variable load and speed fluctuations introduce nonstationary characteristics that impair the reliability of feature extraction and model generalization. To overcome these limitations, this study proposes a novel infrared thermal vision and acoustics-driven multimodal zero-shot diagnosis network (IRT-AMZS). The framework introduces a multireceptive field-based cross-modal fusion (MRCF) module that adaptively extracts and integrates complementary fault features from infrared thermal and acoustic signals, enhancing robustness under varying operational conditions. Furthermore, by employing semantic fault prompts, the IRT-AMZS framework effectively transfers diagnostic knowledge from observed to unobserved fault categories, thereby enabling accurate diagnosis even in the absence of labeled samples for unseen faults. Experimental evaluations on variable-load gearbox datasets demonstrate that IRT-AMZS significantly improves diagnostic accuracy and generalization compared to competing methods.
Accurate prediction of concrete compressive strength is essential for structural safety, material optimization, and cost-efficient construction. While machine learning models have demonstrated strong predictive capabilities, many existing approaches rely on overparameterized architectures, lack physical interpretability, and fail to provide uncertainty estimates required for engineering decision-making. This study proposes a physics-informed and uncertainty-aware stacking ensemble framework for robust prediction of concrete compressive strength. The proposed approach integrates domain-informed feature engineering with monotonicity-constrained gradient boosting models to enforce physically consistent relationships between input variables and strength development. A statistically rigorous out-of-fold stacking strategy is employed to generate unbiased meta-features, which are subsequently combined using an ElasticNet meta-learner to improve generalization. To enhance reliability, the framework incorporates quantile regression for uncertainty estimation, enabling prediction intervals alongside point estimates. Experimental evaluation on the benchmark UCI concrete compressive strength dataset comprising 1030 samples with eight input features demonstrates that the proposed framework achieves competitive predictive performance while significantly improving model robustness and interpretability. Ablation analysis confirms that removing redundant feature transformations and enforcing physics-guided constraints enhance generalization. The proposed method provides a reliable and practically applicable solution for data-driven concrete strength prediction, bridging the gap between advanced machine learning and engineering domain requirements.
This paper investigates the role of viscous damping outrigger trusses in enhancing the seismic performance of structures. Viscous damping outrigger trusses are characterized by high energy dissipation and a stiffness-limited strengthening layer, making them particularly suitable for seismic design of super high-rise structures in high-seismic-intensity regions. The China International Silk Road Center Tower, with a height of 498 m, is currently the tallest building in the world incorporating viscous damping outrigger trusses. To evaluate its seismic performance and the effectiveness of the vibration control measures, a 1:40 scale model shaking table test was conducted. The paper presents the model design, testing process, and key observed phenomena and examines the dynamic response of the structure under 8 frequent, medium, and rare earthquakes. These investigated responses include natural vibration characteristics, dynamic amplification factors, floor displacements, interstory drift ratios, damage evolution, and energy dissipation of the viscous dampers. The test results are in good agreement with finite element analysis results, demonstrating that the structural design is reasonable, the vibration control measures are effective, and the overall seismic performance satisfies the code requirements and predetermined performance objectives. The additional damping ratio provided by the viscous damping outrigger trusses decreases with increasing seismic intensity and further decreases as the main structure enters the elastic–plastic stage. Based on the experimental results, design recommendations are proposed, and the variation trend of the damping ratio is clarified, providing a valuable reference for similar engineering projects.
To prevent the stress of the solenoid valve spool from exceeding the allowable stress and ensure its strength and sealing reliability, the collision process of the solenoid valve spool under the action of electromagnetic force is studied. Firstly, the influences of the force, valve spool stroke, valve seat radius, and ejector rod radius on the valve spool stress are studied by simulation. Then, two mathematical characterizations of the stress on the upper and lower surfaces of the valve spool are obtained, which can better express the dependence of the valve spool stress on the force, valve spool stroke, valve seat radius, and ejector rod radius. Based on this, the valve spool structure is optimized for higher reliability and experimental verification is conducted. Simulation and experimental results show that the original polyimide spool exhibited an upper-surface stress of approximately 160 MPa under repeated impact, exceeding the allowable stress of 120 MPa. After introducing a stainless-steel metal core, the stress transmitted to the polyimide region is reduced to approximately 32 MPa, indicating improved impact resistance and sealing reliability.
In response to the increasingly pressing issue of road traffic noise pollution, this paper draws on the sound absorption mechanisms of porous media to develop a class of composite porous sound-absorbing materials with a graded, adjustable pore configuration and employs COMSOL 6.3 to perform numerical simulations of their acoustic performance. By optimizing key configuration variables such as the composition ratio and arrangement of the matrix, porosity, hydraulic radius of the pores, resonant cavity dimensions, and perforated plate structure, the study systematically analyzed the variation patterns and underlying mechanisms of how these parameters regulate the material’s sound absorption characteristics. The results show that the composition ratio and spatial arrangement of the matrix can significantly improve the material’s noise reduction capability. The low-to-mid-frequency sound absorption performance of the B11 configuration increased by 2.0%–7.0%, while the B25 configuration exhibited the optimal overall sound absorption performance; porosity and pore hydraulic radius significantly influenced sound absorption characteristics. When the porosity was 0.4, the sound absorption coefficient reached as high as 0.999, with a 68.6% improvement in sound absorption across the entire frequency range; increasing the pore hydraulic radius resulted in a maximum increase of 67.1% in the peak sound absorption coefficient. The structural parameters of the perforated plate significantly affect overall sound absorption performance. Among these, optimizing the aperture size can increase the peak sound absorption coefficient by up to 26.9%, while the porosity of the perforated plate has the most pronounced effect on the average sound absorption coefficient, with a maximum increase of up to 176.8%. The graded, adjustable-porosity composite sound-absorbing structure proposed in this paper exhibits excellent broadband sound-absorption characteristics in the critical road traffic noise frequency range of 920–6000 Hz. It can effectively suppress road traffic noise and possesses high engineering practical value and good potential for large-scale application.
Unmanned aerial vehicles (UAVs) with an ultralarge aspect ratio and a joined-wing configuration are light in mass and highly flexible, which makes it challenging to identify their dynamic characteristics underground modal test conditions. In this paper, a full-aircraft ground modal test was conducted for a UAV with an ultralarge aspect ratio and a joined-wing configuration under an approximately free–free suspension condition. A rubber-cord suspension system was adopted to approximate the free–free boundary condition, and the influence of the suspension stiffness was evaluated and corrected for the most affected low-frequency mode. A test model based on the measurement points was established in I-deas for suspension-stiffness correction, and ModalStar2.0 was used for modal data acquisition and phase-resonance tuning. The structural modal parameters were identified using a combined phase separation/phase resonance procedure. Subsequently, a finite element model of the studied UAV was constructed and used for modal analysis. The comparison between the experimental and finite element modal results indicates that the model can reproduce the dominant low-frequency global modal characteristics of the studied aircraft at an engineering level. The results provide experimental modal data and a practical reference for ground modal testing of ultralarge-aspect-ratio joined-wing UAVs.
In order to accelerate the promotion and application of prefabricated bridge constructions in high-intensity areas, this research studied the connection method of prefabricated bridge piers. After clarifying the structural mechanism of the UHPC grouted corrugated duct connection method, the structural parameters were optimized. On this basis, quasistatic experiments and finite-element parametric analysis were conducted. The results showed that the prefabricated pier, which was connected with the optimized UHPC grouted corrugated duct connection method achieved equivalent performance to that of cast-in-place (CIP); optimized construction could reduce engineering costs and improve the ductility of the structure, while the bearing capacity was almost the same as that of CIP piers; the parameterized analysis results have verified the rationality of the proposed structure which can provide reference for engineering design in high-intensity areas.
This study investigated crack characteristics and formation mechanisms in ceramic plates of ceramic/metal composite armor under projectile impact. Finite element models were developed and validated against ceramic/metal beam tests to capture crack initiation and propagation. Within a unified stress wave framework, the impact response was classified into three phases: compressive-shear, spalling, and global bending. Conical and splitting cracks are attributed to tensile waves reflected from the front surface and the evolving conical fracture surface. Spalling cracks are driven by tensile waves reflected from the back surface. At a sufficiently high velocity, comminution initiates when the impact-induced compressive stress exceeds the ceramic's Hugoniot elastic limit. Representative element stress-path analysis further links each crack mode to distinct local stress states and principal stress orientations. Critical impact velocities for conical cracks and comminution were identified based on stress-wave theory, and parametric analyses quantify the effects of projectile velocity, projectile diameter, and ceramic thickness on crack initiation and dominant crack modes. The results provide mechanistic guidance for ceramic/metal composite armor design.
In this research, a smart vibration-based fault diagnosis framework of the automotive suspension system is proposed using multiobjective evolutionary fuzzy classifier (MOEFC). The experimental vibration information gathered was related to a passenger vehicle that is working in a single healthy condition and 10 different fault conditions, which lead to the collection of 1100 samples of the data. It obtained four discriminative factors, root mean square (RMS), kurtosis, dominant frequency and crest factor, to obtain the dynamic properties of suspension behavior. A stratified 10-fold cross-validation approach was taken to provide quality performance evaluation as overfitting was kept to the minimum and the distribution of classes was maintained. Perfectly identified was the healthy condition with an overall classification accuracy of 91.5% in the proposed MOEFC. Receiver operating characteristic (ROC) analysis was performed and resulted in a high class separability with area under the curve (AUC) exceeding 0.90 across all classes of faults. Additional examination showed that dominant frequency and kurtosis would be the greatest contributors to fault discrimination. The developed fuzzy rule-based design offers interpretable decision making, as it allows clear mapping between vibration measurements and fault states. These findings suggest that the suggested solution is a valid and interpretable answer to monitor a suspension condition in controlled experimental conditions.
To efficiently and precisely quantify how the frequency-dependent dynamic characteristics of rail pads affect the fundamental excitation sources of subway ground vibration and wheel-rail rolling noise, a vertical vehicle-rail-fastener coupled dynamic model is constructed via the hybrid spectral element method (SEM) and symplectic method (SM). This model fully utilizes the periodic structural features of rail-fastener components and the propagation law of high-frequency elastic waves in rail members. The proposed method is employed to evaluate how the frequency-dependent characteristics of DT III rail pads-widely used in Chinese subway systems-affect the random dynamic loads, based on experimental measurements. Four main findings are obtained: (1) Under 20 degrees C, the storage stiffness of DT III rail pad presents an approximately logarithmic linear correlation with frequency in the range of 0.1-10,000 Hz. Conversely, its loss factor climbs to a peak value and then declines slowly as frequency increases further. (2) The frequency-dependent properties of rail pads can notably intensify random dynamic loads in two critical frequency ranges: 32-80 Hz (the dominant band for subway environmental vibration) and 500-1250 Hz (the dominant band for wheel-rail rolling noise). Accordingly, it is indispensable to incorporate the frequency-dependent dynamic characteristics of rail pads into the prediction of subway environmental vibrations and wheel-rail noise. (3) Numerically, the effect of rail pads' frequency-dependent behaviors on subway random dynamic loads is comparable to increasing the train speed to 90 km/h. Furthermore, the effect of short-wave track irregularities on subway environmental vibrations is less pronounced than that of rail pads' frequency-dependent characteristics, except when the short-wave irregularities reach an extremely severe level.
Carbon dioxide blasting technology plays a crucial role in current coal-rock mass blasting engineering, predominantly employing bottom energy-release methods. To address the limitations of existing bottom-release approaches, including limited blasting propagation range and susceptibility to "tube ejection," this study proposes a novel bilateral energy-concentrated CO2 blasting technology. The theoretical mechanism of tube wall tearing in double-sided grooved (V-shaped groove) CO2 blasting tubes was first analyzed. Numerical simulations revealed the stress evolution characteristics of four groove configurations (V-shaped, rectangular, semicircular, and trapezoidal) under varying internal pressures and loads. A dimensionless analysis method was adopted to evaluate the influence of groove geometry on plastic limit internal pressure. An experimental system integrating shock pressure monitoring and a high-speed camera was established to characterize shock pressure curves and dynamic jet propagation. The results show that: (1) the stress state adjacent to grooves was higher and mainly in longitudinal tearing tensile failure, with V-groove tubes exhibiting 1.109-1.124 higher ultimate internal pressure resistance than the other bodies; (2) postfracture pressure signatures comprise three sequential phases: exponential escalation, progressive ascent, and oscillatory attenuation, where V-groove specimens achieved a peak impact pressure of 954 kPa, and the shock pressure lasts for 89 mu s; and (3) the CO2 jet from tubes with different groove geometries exhibits similar evolution patterns, characterized by dual jet morphologies ("hemispherical" and "fan-shaped") and four successive expansion stages: ejection, transition, steady diffusion, and dissipation. The jet velocity shows two distinct phases-rapid exponential escalation followed by gradual decline-governed primarily by shock wave and gas expansion effects, respectively. The peak radial velocity varies between 450 m/s and 650 m/s, decreasing in the sequence of V-shaped, rectangular, trapezoidal, and semicircular grooves. The V-shaped blasting tube exhibits superior rupture pattern and jet strength, providing critical empirical guidance for energy-concentrated CO2 blasting tube optimization.
Monopile foundations are widely used in fixed-bottom offshore wind turbines (OWTs), and their seismic response is strongly affected by pile-soil-structure interaction. Most existing studies have focused on OWTs located on flat and idealized seabeds, whereas the influence of near-slope seabed topography has received relatively limited experimental attention. To address this gap, this study conducts small-scale shake table tests on a monopile-supported OWT model embedded in sand under no-slope and near-slope conditions. The effects of turbine orientation angle, toward-slope and away-slope orientations, and the distance from the pile to the slope crest are examined using structural acceleration, soil acceleration, acceleration amplification factors, and pile-head displacement. The results show that, under no-slope conditions, the pile-head and tower-top accelerations increase with the turbine orientation angle, while the soil acceleration is less sensitive to turbine eccentricity. Under near-slope conditions, the measured soil accelerations are approximately 5%-10% higher under the toward-slope orientation than under the away-slope orientation. A shorter distance from the pile to the slope crest generally corresponds to a higher soil acceleration response within the tested range. Compared to previous studies based mainly on flat seabed assumptions or numerical models, this study provides controlled physical model evidence for understanding the seismic response of monopile-supported OWTs near sloping seabeds. The findings provide preliminary experimental support for considering seabed topography effects in the seismic assessment of fixed-bottom OWTs.
Traditional online monitoring systems often rely on high-quality data samples or external expert knowledge, and their model accuracy degrades significantly when equipment states deviate from expected conditions over time. In the preresearch stage of mechanical equipment, researchers face not only data scarcity and severe class imbalance but also performance degradation caused by nonstationary processes and varying operational conditions. To address these challenges, this paper proposes a novel fusion-enhanced model based on local neighborhood standardization and recursive quadratic kernel entropy component analysis. The proposed LNS-RQKECA framework constructs a data-driven vibration signal analysis model for early fault detection in mechanical equipment under multiple load conditions, relying solely on historical data from normal operation without requiring external experience or prior knowledge. The local neighborhood standardization technique transforms the multiload problem into a single-modal problem, enabling effective load-case fusion. To counteract the feature dilution effect induced by sample fusion, the proposed quadratic kernel entropy component analysis employs advanced kernel-based entropy optimization to enhance feature discriminability through deeper data mining. Additionally, a recursive updating mechanism ensures adaptive model adjustment, allowing the system to handle nonstationary load conditions dynamically. This proposed model is validated using both public datasets and real-world operational data. Experimental results demonstrate that under multiload conditions, the proposed model achieves strong generalization capability and nonstationary process adaptability with minimal training data. Compared with the traditional method, the false missing rate and false alarm rate are reduced by 12% and 15%, respectively.
