Hohai University

To enhance the reliability of rotating machinery, cross-domain fault diagnosis becomes vital for detecting faults under unknown operating conditions. However, multi-source domain imbalanced data present significant challenges, as divergent label distributions across domains cause complex domain-class shifts and degrade the performance of cross-domain fault diagnosis. Moreover, diagnostic models often struggle to learn features from minority classes due to label imbalance within each domain, which may degrade the performance in diagnosing these minority classes. To address these challenges, we propose a Dual-Reweighted Siamese Contrastive learning network (DRSC) for cross-domain fault diagnosis with multi-source domain imbalanced data. In DRSC, we design a Siamese feature extractor based on a wide-kernel convolutional neural network to capture short-term characteristics and leverage the convenience in extracting domain-invariant features. Subsequently, to alleviate domain-class shifts, we design a reweighted contrastive domain-class alignment mechanism that strategically pulls domain-class pairs together while pushing other health conditions away. Finally, to enable the diagnostic model to learn from minority health conditions, a reweighted health condition classifier is developed by assigning higher weights to the minority classes. Evaluation results on two public datasets illustrate DRSC outperforms comparison models in cross-domain fault diagnosis.

Piezo‐catalysis has been widely studied for water environment remediation, but still faces weak intrinsic piezo‐response and lack of active sites for generating reactive oxygen species. In this study, phosphorus‐doped BiOCl (PBOC) is rationally designed for efficient piezo‐catalytic degradation of bisphenol B (BPB). Characterization results reveal that phosphorus is doped in the lattice by chlorine substitution, accompanied by the generation of a substantial amount of oxygen vacancies (OVs). This enhances the material's molecular dipole, resulting in increased intrinsic piezoelectricity. Simultaneously, the production of reactive oxygen species (ROS), including hydroxyl and superoxide radicals, is significantly higher than BOC. The piezo‐catalytic degradation rate of BPB by PBOC increases to 0.179 min⁻¹, which is 7.8 times that of BOC. Density functional theoretical (DFT) calculations reveal that the doped P and the simultaneous generation of OVs not only enhance the molecular dipole, but also serve as active sites for adsorbing and activating O2 and H2O to efficiently generate superoxide radicals and hydroxyl radicals, respectively. This work demonstrates a simple but efficient approach to promote piezo‐catalytic environment remediation.

Particle finite element method (PFEM) can effectively simulate large deformation problems in geotechnical disasters such as landslides, debris flows, and dam breaks. In recent years, PFEM has attracted much attention at home and abroad. The research progress of PFEM for large deformation simulation in geotechnical engineering is reviewed. Firstly, the development history and basic idea of the PFEM are introduced. Then, the theoretical progress of the computational theory for PFEM in geotechnical engineering is presented. Finally, the application progress of the PFEM for large deformation simulation in geotechnical engineering is introduced, including collapse and landslide problems, structure–soil coupling large deformation problems, hydromechanical coupled problems, etc. Through the review of the research progress of PFEM for large deformation simulation in geotechnical engineering, the cognition of relevant researchers in this field is deepened, and the development of large deformation simulation theory and engineering application of PFEM for geotechnical engineering is promoted.

Water shortages are intensifying globally due to climate change and human activities. Northeast Asia, with diverse ecosystems and transboundary water systems, is particularly sensitive to these pressures. Yet, the region’s water resource changes and drivers remain largely unknown. Here, we integrate Landsat and Sentinel-2 images, Gravity Recovery and Climate Experiment and its Follow-On observations, climate and anthropogenic data, finding a net surface water area loss of 16 × 103 km² in Far East Russia over 2000−2023, primarily driven by rising temperature and evaporative demand, and a net surface water area gain of 3 × 103 km² in Northeast China, primarily driven by increasing precipitation and irrigation infrastructure. Approximately 1004 0.5° gridcells (1.4 × 106 km²) have concurrent losses of surface water area and total water storage. Approximately 185 million people reside in watersheds with surface water area or total water storage loss, underscoring the need for sustainable water management under intensifying climate change and human activities.

Owing to the excellent capability in dealing with label ambiguity, Label Distribution Learning (LDL), as an emerging machine learning paradigm, has received extensive research in recent years. Though remarkable progress has been achieved in various tasks, one limitation with existing LDL methods is that they are all based on the i.i.d. assumption that training and test data are identically and independently distributed. As a result, they suffer obvious performance degradation and are no longer applicable when tested in out-of-distribution scenarios, which severely limits the application of LDL in many tasks. In this paper, we identify and investigate the Generalizable Label Distribution Learning (GLDL) problem. To handle such a challenging problem, we delve into the characteristics of GLDL and find that the label annotations changing with the variability of the domains is the underlying reason for the performance degradation of the existing methods. Inspired by this observation, we explore domain-invariant feature-label correlation information to reduce the impact of label annotations changing with domains and propose two practical methods. Extensive experiments verify the superior performance of the proposed methods. Our work fills the gap in benchmarks and techniques for practical GLDL problems.

Improving our understanding of terrestrial carbon‐water coupling processes at the basin scale is crucial for assessing the impacts of global change on river basin's ecosystem services. Supported by remote sensing technology, previous studies have investigated the variability of carbon sequestration or water yield of a basin separately. However, the integration of these two components in an assessment system has rarely been considered, which may lead to partial appraisals of eco‐hydrological services. To address this knowledge gap, we have developed a new remotely sensed index, the Carbon‐Water Efficiency Index (CWEI), which is derived as the product of the normalized precipitation use efficiency (PUE) and water yield efficiency (WYE) during a specified period for a given basin, to quantify the coupling carbon‐water ecosystem service. In this study, the spatial and temporal variabilities of CWEI were investigated using satellite‐based products across the 38 major river basins from 1982 to 2016. Furthermore, the response patterns of CWEI to its components were explored, and the contributions of the three variables (i.e., temperature, precipitation, vegetation coverage) on changes in CWEI were quantified for each of the basins. The results demonstrated that the CWEI exhibited an increase in 31.6% of the world's major river basins, of which 83.3% with an increasing CWEI demonstrated a response pattern of increasing PUE and decreasing WYE. Furthermore, temperature was identified as the primary contributor to CWEI in 50% of the basins. It is noteworthy that green hotspot basins, where afforestation projects were underway, exhibited enhancements in reduced water yield pressure and demonstrable benefits to local ecohydrology. The proposed CWEI provides a new framework for comprehending the sustainable development of river basins worldwide.

The floating offshore wind turbine (FOWT) undergoes six-degree-of-freedom motions influenced by wind, waves, and currents. Surge and sway motions influence the aerodynamic performance of the FOWT by creating variations around the blade flow field, leading to complex dynamic stall mechanisms of the airfoil. This study examines the S809 airfoil, coupling pitching motion with surge and sway two-degree-of-freedom motions in both in-phase and anti-phase scenarios. The study examines the impact of combined surge, sway, and pitching motions on hysteresis response characteristics, surface vortex separation, and load variations during dynamic stall. The results indicate that in-phase motions forward swaying and downward surging significantly enhance lift but induce severe unsteady flow characteristics, increase separation vortex intensity, and hamper flow field recovery. Conversely, backward swaying and upward surging reduce boundary layer kinetic energy and turbulence, decreasing lift and stall severity. In anti-phase motions, forward swaying and upward surging increase boundary layer instability, while backward swaying and downward surging weaken boundary layer energy and enhance disturbances, delaying stall. These findings provide theoretical support for FOWT safety and stability, offering insights into wind turbine design, optimization, and operation.

Automatic detection of asphalt pavement cracks on digital images is crucial for pavement condition assessment. However, pavement images are prone to blurring due to noise generated by imaging devices, which undermines the ability of semantic segmentation models to extract features. Current denoising algorithms show limited capability in handling real noise in pavement images, while most semantic segmentation models struggle to detect fine cracks effectively and accurately. Aiming to tackle these challenges, we propose an integrated denoising and segmentation network (IDSNet), which integrates adaptive denoising filters into a semantic segmentation network. The proposed IDSNet comprises the multi-filter denoising module, hyperparameter prediction module, and semantic segmentation module. The multi-filter denoising module is introduced to suppress background noises and shadow interference. The hyperparameter prediction module is employed to tune the filter parameters based on the loss of the semantic segmentation module. Additionally, a spatial and channel reconstruction convolution operation is embedded into the Attention U-Net to enhance its ability to extract features of fine cracks from context information. Experimental results demonstrate that IDSNet outperforms other state-of-the-art segmentation models in detecting asphalt pavement cracks, achieving a mIoU of 55.10% and an F1-Score of 60.86% on the private Asphalt Pavement Crack Denoising Image Dataset (APCD dataset). Compared to the traditional Attention U-Net, IDSNet demonstrates improvements of 6.88% in mIoU and 6.51% in F1-Score, respectively. Accordingly, IDSNet will assist in image denoising and intelligent crack detection for asphalt pavement.

Landslides pose significant threats to infrastructure and human safety, necessitating advanced monitoring and early warning technologies. This study simulates the landslide initiation mechanisms through laboratory-scale sand column collapse experiments. An innovative integration of dynamic optical frequency domain reflectometry (OFDR) with particle image velocimetry (PIV) is presented to achieve high-resolution synchronous monitoring of surface velocity fields and internal strain distribution during the collapse. Micro-disc anchor plates were added along the fiber optic cables to optimize the mechanical coupling between the cable and soil, the response accuracy and reliability of distributed fiber optic sensing (DFOS) to rapid deformation were significantly improved. Testing results revealed the four-stage dynamic evolution pattern of sand column collapse: vertical collapse, transition from vertical to horizontal flow, horizontal flow, and deceleration and cessation. Based on the OFDR strain data, the formation and propagation mechanisms of shear bands in each stage were clarified. The surface velocity field captured by PIV was highly consistent with the strain distribution monitored by DFOS, validating the feasibility of multi-source data fusion analysis. The results show that micro-anchored fiber optic cables effectively identified the progressive failure from shallow to deep soil mass. The “sine wave” characteristic of strain is greatly related to the location of the shear slip surface, providing new insights into the internal deformation mechanisms of landslides. Through technological integration and methodological innovation, the limitations of temporal and spatial resolution of traditional monitoring methods were overcome, offering reference for the early identification of landslide disasters and the study of their dynamic theories.

The unprecedented demographic shifts toward an aging population pose significant challenges to global healthcare systems. Understanding the heterogeneity in disease prevalence among the elderly is crucial for effective public health strategies. Using prevalence data of 85 types of age‐related diseases, we calculated the global heterogeneity of disease distribution by the Shannon Diversity Index (SHDI). We observed significant geographic variations in disease heterogeneity, with higher SHDI values in high‐income Western countries such as the United States of America and Sweden and lower in South Asia and Oceania (p < 0.05). In 2021, SHDI values in elderly populations (age ≥60 years) for Europe and North America countries were an average of 1.12 times higher than in Oceania. While SHDI increases toward higher ages (for instance, in 2021, SHDI for adults above 95 years is 1.06 times higher than for ages between 60 and 64 years), the global SHDI tends to decrease nonlinearly over time. From 1990 to 2021, global age‐standardized SHDI (age ≥60 years) averagely decreased by 1.2% for both men and women. Our analysis further revealed that socio‐economic factors (e.g., socio‐demographic indices, governance) strongly impacted global SHDI changes, while climatic and environmental factors (e.g., extreme climate and air pollution) showed significant differences across genders. Our study highlights the need for implementing comprehensive healthcare strategies, focusing on reducing health disparities and addressing environmental and socio‐economic determinants to address inequalities in age‐related diseases effectively.

Physically based urban wash‐off modeling presents a promising approach for investigating the dynamics of road‐deposited sediments (RDS) and the associated pollutants during rainfall events. This paper proposes a novel physically based model to predict urban wash‐off process over impervious surfaces, where raindrop‐induced detachment, flow‐driven detachment and deposition are computed separately. This is achieved by modifying the Hairsine‐Rose (HR) model to account for sediment trapping due to road roughness, and incorporating multi‐sized particles to capture shielding effects during detachment and particle‐size selection during deposition across varying particle size distributions (PSDs). The model is validated through laboratory experiments, including single‐size particle simulations under varying rainfall conditions and particle sizes, and multi‐size particle simulations with different PSDs. Numerical experiments are conducted to systematically examine the relationships between model parameters and influencing factors (i.e., rainfall intensity and particle size). The results enhance parameter interpretation and simplify the model. Model predictions show strong agreement with experimental measurements, demonstrating that model parameters correlate with respective influencing factors and remain physically interpretable. By incorporating multi‐sized particles to reflect the PSD, the model effectively captures the particle‐size selection phenomenon during detachment and deposition. This study provides new insights into RDS wash‐off process modeling.

The widespread use of low-heat Portland cement in hydraulic concrete highlights the significance of its sulfate-resistant quality in engineering. In this study, the porosity, sulfate ion concentration, compressive strength, Vickers hardness and linear expansion of low-heat cement paste were measured under the conditions of 0.5%, 5.0%, and 8.0% solution concentrations of Na2SO4 and 5.0% solution concentrations of MgSO4, and (NH4)2SO4. The phase composition and microstructure of the eroded cement pastes were measured using X-ray diffraction, thermogravimetric analysis, and scanning electronic microscopy. The results show that the linear expansion rate and compressive strength loss rate of cement paste increase with the increase of sulfate solution concentration. However, the relationship between sulfate solution concentration and sulfate ion penetration depth is not completely proportional. In the lower concentration of sulfate solution, sulfate ions show stronger ion transport ability in the cement slurry. After 360 days of exposure to sulfates, the compressive strength loss rate of low heat cement in magnesium sulfate solution and ammonium sulfate solution was 3.1 times and 7.2 times that of sodium sulfate solution. The results indicate that, in sodium sulfate solution, the sulfate resistance of the specimens is strongest, followed by magnesium sulfate and ammonium sulfate.

This study analyzes key meteorological parameters and aerosol classifications over the LHAASO region from 2021 to 2023 using satellite-based remote sensing. The meteorological parameters, average temperature (Tavg), relative humidity (RH), wind speed (WS), and precipitation, play a vital role in influencing aerosol behavior, distribution, and transformation processes. Data were sourced from Land Data Assimilation System FLDAS, the Global Land Assimilation System GLDAS model, and Aerosol in Some Contexts AIRS, enabling a comprehensive monthly and annual analysis of atmospheric conditions. Among the aerosol types examined, Dust Column Mass Density consistently dominated the aerosol composition, ranging from 70.34 to 82.66%. Notable interannual variations were observed in Carbon Column Mass Density, while Sulfate Emission and SO₄ Surface Mass Concentration remained relatively low, between 0.01% and 2%. Biomass burning demonstrated substantial yearly variation, contributing 56.31%, 16.55%, and 37.24% over the study period. Key findings include a positive correlation between Tavg and precipitation (0.768), Tavg and RH (0.947), and a negative correlation between Tavg and WS (-0.741), impacting aerosol distribution and transformation. The Generalized Additive Model (GAM) shows that Tavg significantly influences SO₄, Dust, and Biomass, with WS affecting SO₄. Precipitation and RH significantly impact Dust (precipitation: p = 0.00663, RH: p = 0.00223), with RH showing a marginal influence on Biomass (p = 0.0645). Integrating satellite-based observations with meteorological data, this study improves understanding of aerosol variability and its potential impacts on regional climate and air quality, offering perspectives for future atmospheric research and environmental monitoring in high-altitude environments.

Understanding the three-dimensional in-situ stress field is crucial for estimating the stability of large deep underground powerhouse. However, due to the insufficient representativeness and unreliability of in-situ stress measurements, it is difficult to determine the complete 3D in-situ stress field around large deep underground powerhouse based on limited in-situ stress data points. This study focuses on the underground powerhouse of the GX hydropower station. After introducing the geological conditions and topography of the underground powerhouse area, the in-situ stress test results are interpreted to clarify the distribution characteristics of natural stresses in the project area. Notably, in the region of the large underground powerhouse area, the horizontal major principal stress ranges from 16.81 to 32.64 MPa, and the horizontal minor principal stress ranges from 9.35 to 18.22 MPa, indicating relatively high stress levels. The measured principal stress in the underground powerhouse is generally oriented towards the NE, with the in-situ stress being primarily dominated by horizontal tectonic stress. Additionally, a new in-situ stress field calculation framework was proposed, which was used to invers the in-situ stress in the underground powerhouse area. A noteworthy finding is that topography and geological structures are the main factors influencing the distribution of in-situ stress. The distribution of principal stresses is generally proportional to depth, with principal stresses increasing to varying degrees as depth increases. However, in local fault zones, stress release occurs, and the in-situ stress field is significantly segmented.

In real-world Industrial Internet of Things (IIoT) scenarios, due to the limited storage capacity of IIoT devices, fresh data continuously received by diverse devices will overwrite the outdated data and change the local data distribution. However, state-of-the-art studies have demonstrated that Federated Learning (FL) tends to focus on training with fresh data, and the latest global model may forget the historical update directions (i.e., catastrophic forgetting). This issue can significantly degrade the global model accuracy. Existing methods primarily focus on integrating outdated data characteristics into fresh data but overlook the large parameter update gap between global and local models during global aggregation. This gap can cause the global model updates to deviate from the optimal direction. To this end, we propose a federated adaptive weighted aggregation method based on model consistency (FedAWAC). Specifically, FedAWAC measures the model consistency on devices and dynamically adjusts the aggregation weights of each local model, thereby guiding the global model toward optimal updates. Furthermore, FedAWAC integrates M historical global models most correlated to the latest global model on the cloud server to overcome catastrophic forgetting. Experiments on 4 different datasets (Non-IID settings) indicate that compared to 5 baselines, FedAWAC can improve global model accuracy by an average of 1.86%, reduce the forgetting rate by an average of 3.91%, and save average memory usage by up to 2.57GB.

Land Surface Temperature (LST) is an important parameter representing surface energy, which is of great significance for monitoring urban heat islands, agricultural drought, and global climate. The high-resolution LST observations will address new applications in hydrology. The spatiotemporal fusion method can generate LST with high temporal and spatial resolution. However, missing data due to cloud cover becomes a main limitation to improving the accuracy of spatiotemporal fusion models. The purpose of the fusion of LST is image prediction and generation, and deep learning generative models provide an effective idea to solve this problem. Therefore, we proposed a conditional generative model fusion network (CGMFN) for LST generation in this paper. Firstly, based on the generated model, we construct an unsupervised generation network that simultaneously learns and iterates, which can generate fine-spatiotemporal-resolution LST data from reference images with missing values. Then, spectral normalization was applied to generators and discriminators to stabilize the training process. The pre-training mechanism was adopted to improve the iteration efficiency of the model. We tested and evaluated the model in the Heihe River Basin using FY-4A LST and MODIS LST datasets. Compared to different methods, CGMFN produces a lower RMSE (average<0.4K), LPIPS (average<0.11), and higher SSIM (average>0.97). In practical applications, CGMFN can reduce the influence of reference image missing values on fusion results and generate land surface temperature products with reliable accuracy.