For computational wind engineering applications, selecting suitable synthetic turbulent inflow generation methods remains challenging due to their diversity and lack of standardized recommendations. Motivated by findings from the companion paper (Waleson et al., 202x), this study compares five approaches – DFSR, PRFG3, Mann, TurbSim, and DFSEM – under identical large-eddy simulation setups to reproduce atmospheric boundary layers for open, suburban, and urban terrains based on ESDU reference profiles. The comparison reveals distinct performance trends: DFSR achieves the closest match to prescribed turbulence statistics through customizable input spectra suited to non-standard target conditions; PRFG3 offers comparable accuracy with greater adaptability regarding mesh or time-step variations; Mann reproduces Reynolds shear stresses but requires further adaptation for wind engineering applications; TurbSim shares similar limitations with fewer options to ensure anisotropy; and DFSEM enables direct stress specification when sufficient upstream space is available. These results offer practical selection criteria among existing methods and establish a foundation for standardized LES inflow generation practices in CWE.
Pollutant dispersion from vessel emissions is investigated in an atmospheric boundary layer wind tunnel using a generic geometry representative of an offshore service vessel. The vessel model is exposed to neutral maritime flow conditions and densimetric Froude number similarity to the real-world configuration is enforced to realistically replicate buoyancy-driven plume dynamics. Time-averaged concentration levels are assessed at 1.5 m above deck for three wind directions, i.e., γ = 0° (headwind) and γ = ±15°, two Maximum Continuous Rating (MCR) engine loads (25% and 75%), and two approach flow velocities (10 m/s and 20 m/s at exhaust height). An analytical framework is proposed to relate measured concentration fields to potential exposure levels. Long-term and short-term exposure levels are investigated for the most critical wind direction (γ = −15°). Results demonstrate that deck-level exposure is governed by the interaction between buoyant plume dynamics, vessel-induced wake structures, and exhaust momentum. For γ = 0° and −15°, plume–wake interaction causes pollutant trapping and elevated deck-level concentrations, whereas for γ = 15° exhaust momentum and wind direction transport pollutants away from the vessel. Increasing engine load enhances exhaust momentum, enabling the plume to escape recirculation regions and reducing on-deck concentrations. It is further shown that hazardous gas exposure can exceed occupational limits, but exhaust gas after-treatment systems and low-sulfur fuel effectively reduce exposures, enhancing occupational safety and mitigating potential health impacts on deck. Overall, this work shows that controlled wind-tunnel measurements combined with exposure-oriented analysis can translate vessel-scale dispersion behavior into safety-relevant metrics.
The shedding of spanwise vortices of a streamlined single box girder is the primary cause of vortex-induced vibration (VIV). The three-dimensional (3D) spanwise perturbation method suppresses VIV by exciting the secondary instability of the spanwise vortex, promoting the transition of the spanwise vortex into streamwise vortices. The Internal Suction-Blowing Method (ISBM) achieves flow control by periodically arranging suction and blowing holes on the girder surface. This study validates the effectiveness of the ISBM in controlling the VIV of a single box girder through wind tunnel tests. The results indicate that the ISBM is an efficient flow control method without negative effects. The optimal perturbation position is the trailing edge of the girder underside. Smaller perturbation spacing, larger single-hole flow rate coefficients, and larger hole sizes yield the best control effects. The minimum effective single-hole flow rate coefficients are CQv = 2.28 × 10−3 and CQt = 0.97 × 10−3 for vertical and torsional VIV, respectively. The minimum effective hole size is a = 3 mm, and the optimal dimensionless spanwise perturbation spacing is λ = 2. Wake analysis results demonstrate that the ISBM suppresses the VIV of the single box girder by exciting the three-dimensional instability of the spanwise vortex.
The increase of bridge span can lead to an overlap between the vertical and torsional lock-in regions, potentially resulting in a more complex dynamic behavior than that of solely vertical or torsional vortex-induced vibration (VIV). To explore this issue, a refined sectional model in wind tunnel with adjustable vertical-torsional frequency ratio and Scruton number ratio is firstly designed. Then, dynamic properties of the two degrees-of-freedom (DOF) VIV system, including amplitude branch, mode switching, and system stability, are systematically investigated. And the influences of frequency and Scruton number ratios are examined. Finally, through surface fluctuating wind velocity measurement, flow field visualization, and flow field disturbance, the triggering mechanism and evolution process of the two DOF VIV are revealed. The results show that, with the variation of vertical-torsional frequency ratio, the vertical and torsional VIVs can experience coupling and competition phenomena, which are accompanied by a mode switching process. The dynamic properties of the two DOF VIV system are less sensitive to the initial perturbations but significantly influenced by the Scruton number ratio. The vertical and coupled vertical-torsional VIVs are primarily driven by the Kármán vortex street, while the torsional VIV is jointly driven by leading-edge and trailing-edge vortex shedding.
As key components of long-span cable-stayed bridges, parallel stay cables are significantly affected by both Reynolds number effects and aerodynamic interference. To clarify their aerodynamic force characteristics in the subcritical and critical Reynolds number regions, wind tunnel tests were carried out to measure the mean and fluctuating drag/lift coefficients, pressure coefficients and Strouhal numbers under various arrangements. The results show that the upstream cable presents a typical drag crisis in the critical region, and its aerodynamic characteristics are slightly interfered only at small spacings. A sharp increase in fluctuating lift in the critical region can lead to significant wind-induced vibrations. For the downstream cable, its mean and fluctuating aerodynamic forces are dominated by the Reynolds number and spacing coupling: as the cross-wind spacing increases, the trend is similar to that of a single cable. The Strouhal number fluctuates greatly in the subcritical region and stabilizes in the critical region. The separation bubble and boundary layer transition of the downstream cable are controlled by a clear spacing threshold, and the asymmetric circumferential pressure distribution directly leads to abrupt aerodynamic variations in the critical region. This study provides a reliable reference for the wind-resistant design of parallel stay cables.
Twin-box girders generally exhibit favorable flutter stability. However, their distinctive aerodynamic profile makes them susceptible to vortex-induced vibration (VIV). The effects of central gap width and maintenance-track position on the VIV performance of a twin-box bridge girder were investigated through wind tunnel tests. Twenty cases combining five gap-width ratios (G/W = 0.07–0.29) and four normalized track positions (x/B = 0.16–0.91) were tested. VIV responses and surface pressures were measured simultaneously. Proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) were applied to the fluctuating pressure field. The VIV responses were strongly influenced by both the gap width and the maintenance-track position, with narrower gaps and more inward track positions generally resulting in lower VIV amplitudes. Among the tested cases, the narrowest gap (G/W = 0.07) exhibited the lowest overall VIV responses, whereas the maximum vertical and torsional responses occurred at G/W = 0.24 and G/W = 0.29, respectively. The most unfavorable maintenance-track positions were x/B = 0.63 for vertical VIV and x/B = 0.42 for torsional VIV. Pressure and modal analyses identified the inner inclined web and upper surface of the downstream girder as the regions with large pressure fluctuations. Low-order POD/DMD modes captured most of the fluctuating-pressure energy, and the first-mode energy ratios varied consistently with the VIV amplitude.
Developments in dynamic airtightness evaluation methodologies are pivotal for the airtightness design and passenger comfort of railway trains. However, existing methods exhibit notable shortcomings when assessing trains traversing tunnels. To address this limitation, this study proposes a novel definition of dynamic airtightness and its corresponding evaluation methodology based on field test pressure data. Results revealed that the newly-proposed local dynamic airtightness provides a better approximation of field test data compared to the conventional global definition. The piecewise linear representation (PLR) algorithm based on perceptually important points (PIPs) was introduced into the local dynamic airtightness evaluation. Through comparative analysis, the effectiveness of the proposed method was validated, and the rational design of key evaluation parameters was investigated. Furthermore, invalid pressure segments were identified and eliminated to minimize evaluation errors. Compared to the conventional global definition, the equivalent global dynamic airtightness derived from the local definition provides a more accurate description of the actual overall airtightness performance. The pressure waves were characterized by the characteristic segment model: PW=1*In1+m*CWi+n*EWi+1*Out1. On this basis, the theoretical determination model for primary turning points of pressure waves was established as: S=5+2*(N1′+N2′).Based on this, the adaptability of the evaluation method to pressure waves with different characteristics was developed for the local dynamic airtightness. These findings provide important methodological support for the research and design of train airtightness.
Large-eddy simulation (LES) of urban wind flows is often assessed using nominal grid spacing, but the smallest motions resolved are strongly influenced by both subgrid-scale (SGS) modelling and numerical dissipation. A clear research gap is that the SGS dependence of effective resolution has not been quantified systematically for isolated-building benchmarks. This study aims to quantify how commonly used SGS models modify the effective cut-off length scale in LES of the airflow around a 1:1:2 isolated building. A set of OpenFOAM simulations was performed on an identical grid using several SGS models and an upwind-blended advection setting for reference. The effective cut-off scale was identified by matching LES velocity spectra to wind-tunnel spectra after applying a simple low-pass filtering procedure and then converted to an equivalent length scale based on the local mean convection velocity. The most dissipative configurations showed the earliest decay of high-frequency energy and the largest cut-off scales, particularly in the roof shear layer and wake. In contrast, dynamic and structure-sensitive models produced smaller cut-off scales and more compact spatial distributions, indicating improved effective resolution on the same mesh. These findings imply that LES quality control for urban aerodynamics should complement nominal grid checks with effective-resolution metrics, and that overly dissipative combinations of numerics and SGS modelling should be avoided when spectral fidelity and unsteady wind statistics are key targets.
A numerical simulation method is proposed based on ANSYS Fluent to analyze the control performance of underwater heave plates (UHPs) for mitigating the flutter of long-span bridges. Compiled user-defined functions (UDFs) are developed to account for wind–deck interaction, water–heave plate interaction, and deck–heave plate coupled motion. The method is validated against wind tunnel and water tank experimental results. It enables safe and reliable acquisition of the time-domain responses of the rigid deck–cable–UHP coupled system, as well as the evolution of the surrounding flow field, providing a solid basis for investigating the application of UHPs in bridge flutter mitigation. The coupled motion behavior and energy dissipation process of the system are analyzed, and the optimization principles governing UHP mass are clarified. Finally, case studies on three decks with different cross-section confirm that UHPs achieve consistently effective flutter suppression, providing practical references for decks with similar configurations. These findings provide valuable insights and technical guidance for applying UHPs in the wind-resistant design of long-span bridges.
In recent years, many long-span parallel bridges have been built worldwide, replacing existing structures and meeting the ever-increasing demands of vehicle traffic. However, their complex aerodynamic interference effects and vortex-induced vibration (VIV) problems are becoming more prominent. In this study, the section models were made according to a cable-stayed bridge with twin-separate parallel Π-shaped decks. Then, the L-shaped deflectors with different sizes were installed on the section models, and synchronous pressure- and vibration-measured tests were performed to study the influence of the horizontal length, vertical height, and layout of L-shaped deflectors on the VIV performance. Moreover, the sizes and layouts of L-shaped deflectors under the present experimental conditions, which can effectively suppress the VIVs of the twin-separate parallel Π-shaped decks, were proposed. Furthermore, numerical wind tunnel simulations were also performed to investigate the VIV suppression mechanism of L-shaped deflectors on the twin-separate parallel Π-shaped decks. Therefore, this study can provide a meaningful reference for the VIV suppression of similar bridges.
Understanding how upstream terrain roughness influences flow around bluff bodies is essential for predicting flow features relevant to wind loading and future ventilation or pollutant-dispersion studies in urban environments. This study combines wind tunnel experiment and numerical simulations to examine how explicitly resolving upstream terrain roughness and floor-mounted obstacles used to generate an atmospheric boundary layer (ABL) affect the flow around a cubical building model. Large Eddy Simulations (LES) through the Lattice Boltzmann Method (LBM) with the Immersed Boundary Method (IBM), and the Synthetic Eddy Method (SEM) is validated against Particle Image Velocimetry (PIV) measurements from the Boundary Layer Wind Tunnel at the University of Bristol. Four numerical upstream terrain configurations, S0, R1, R2 and R3 were analysed using different relative roughness-height ratios, k/h, where k is the roughness-element height and h is the building height, which R1 case replicates the experimental ABL conditions. The flow response was examined across five regions, namely, approaching flow, windward stagnation, rooftop separation, near wake, and downstream recovery. Results show that resolving upstream roughness significantly improves agreement between LES and experiments, particularly in vertical velocity profiles, stagnation-point location, and turbulence statistics near the building. Increasing k/h enhances turbulence generation and vertical mixing in the incoming ABL, steepens near-ground shear, and reduces both the separation bubble thickness and the near-wake recirculation width. In addition, the predictive fidelity of a truncated LES domain was assessed against the roughness-resolved simulations and wind-tunnel measurements, highlighting the extent to which reduced-domain set-ups can reproduce key mean-flow, turbulence, and loading-relevant metrics.
Flow field compression is a key milestone for the feasible implementation of deep learning (DL) models at scale and the efficient storage of the flow datasets required for their training. Sampling strategies represent a fundamental step in vision-based compression techniques, directly conditioning both model accuracy and compression ratios. Their main objective is to accurately capture all necessary flow features for reproducing the phenomena of interest, which typically involves extracting wind-induced forces while simultaneously resolving flow characteristics in the near and far wakes. However, accurately identifying flow information in those regions requires conflicting sampling criteria. To address this challenge, this study proposes an importance sampling strategy guided by flow activity to automatically identify and focus on high-activity regions within the flow domain, combining signed distance functions (SDFs) and vorticity fields. The proposed flow activity-biased importance sampling (FABaS) method achieves near-lossless compression ratios of ∼37718:1 while guaranteeing accurate reproduction of flow features, with a mean absolute percentage error (MAPE) of 0.063%, for precise surface force extraction in the boundary layer region and far wake. Consequently, this technique is a powerful tool for deep learning compression and prediction of flow field data. The FABaS code is available on GitHub.
As long-span parallel bridges are increasingly constructed, aerodynamic interference between adjacent decks warrants special attention. However, its effects on wind-train-bridge (WTB) responses and train running safety remain unclear. Taking the Quanzhou Bay Bridge as the engineering background, this study develops a time-domain WTB coupled framework considering adjacent-bridge aerodynamic interference. Aerodynamic parameters of the train-bridge system are obtained by computational fluid dynamics (CFD), while bridge aerodynamic forces are evaluated using bridge wind engineering theory incorporating the Y. K. Lin model. The rate of change (ROC) is further introduced in the wind-speed–train-speed parameter space to quantify interference effects. Results show that the interference is most pronounced when the railway bridge is on the leeward side. Because the incoming-flow aerodynamic moment and the torque induced by the eccentric train load act in the same direction, the midspan torsional displacement increases by 214% relative to the no-interference case. Train vertical responses are weakly affected, whereas lateral responses and safety indices exhibit greater sensitivity to aerodynamic interference; under strong-wind and high-speed conditions, the ROCs in the derailment and overturning coefficients of the head car reach up to approximately 30% and 25%, respectively. During train passage, the vertical self-excited load accounts for approximately 15%–25% of the total vertical aerodynamic load acting on the bridge, whereas the corresponding torsional contribution is generally below 5%. The Y. K. Lin and quasi-steady formulations yield differences below 5% for most indices and operating conditions; however, the relative difference in the bridge midspan vertical acceleration locally reaches 16%.
The number of modern low-rise buildings continues to grow, and sloped roofs are susceptible to wind-induced damage due to airflow separation and significant wind pressure fluctuations. Accurate prediction of wind pressure coefficient statistics for sloped roofs is important for wind-resistant design. However, dense wind-direction sampling is difficult in engineering applications, and large intervals cause insufficient target-domain samples, reducing the accuracy of data-driven models. To address this issue, this paper proposes a transfer learning framework for improving the prediction accuracy of wind pressure coefficient statistics for low-rise sloped-roof buildings under sparse wind-direction sampling. Under a unified experimental setup, Plain Convolutional Neural Network, Residual Network, Multilayer Perceptron, and Extreme Gradient Boosting were constructed and compared for predicting the mean, root mean square, maximum, and minimum wind pressure coefficients. The best-performing model was trained with source-domain data to obtain a pretrained model. Before target-domain fine-tuning, sparse sampling schemes at 15° and 25° were examined to analyze the effect of wind-direction sparsity and determine the fine-tuning interval. The model was then fine-tuned using limited target-domain samples and compared with a model directly trained on the same samples, including univariate cases with eave-height variations and bivariate cases with simultaneous variations in eave height and roof slope. The results indicate that transfer learning improves prediction accuracy under sparse target-domain samples. Finally, Shapley Additive Explanations and Partial Dependence Plot analyses were conducted to enhance interpretability and improve the potential engineering application value of the proposed framework.
