<rss version="2.0" xmlns:atom="https://www.w3.org/2005/Atom">
  <channel>
    <title>Transport Research International Documentation (TRID)</title>
    <link>https://trid.trb.org/</link>
    <atom:link href="https://trid.trb.org/Record/RSS?s=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" rel="self" type="application/rss+xml" />
    <description></description>
    <language>en-us</language>
    <copyright>Copyright © 2026. National Academy of Sciences. All rights reserved.</copyright>
    <docs>http://blogs.law.harvard.edu/tech/rss</docs>
    <managingEditor>tris-trb@nas.edu (Bill McLeod)</managingEditor>
    <webMaster>tris-trb@nas.edu (Bill McLeod)</webMaster>
    <image>
      <title>Transport Research International Documentation (TRID)</title>
      <url>https://trid.trb.org/Images/PageHeader-wTitle.jpg</url>
      <link>https://trid.trb.org/</link>
    </image>
    <item>
      <title>Digital twin-based non-stationary wind field reconstruction with time-varying coherences for long-span bridges</title>
      <link>https://trid.trb.org/View/2670154</link>
      <description><![CDATA[Long-span bridges in mountainous regions are often subjected to non-stationary winds. Therefore, the non-stationary wind fields need to know in performing the buffeting analysis of these bridges. However, the limited anemometers equipped on a long-span bridge are insufficient to obtain a complete non-stationary wind field. This paper thus proposes a digital twin-based method to reconstruct a complete non-stationary wind field with time-varying coherence for a long-span bridge in a mountainous region. The real non-stationary wind field and the non-stationary wind characteristics measured by limited anemometers are taken as a physical entity. The conditionally simulated wind field is taken as a virtual entity. By mapping the non-stationary wind characteristics measured by limited anemometers to the conditionally simulated non-stationary wind field and at the same time by optimizing the time-varying coherence function, a complete non-stationary wind field along the bridge deck can be reconstructed. The proposed method is applied to a real long-span suspension bridge in a mountainous region. The results demonstrate that the reconstructed non-stationary wind field not only matches with the measured fluctuating wind speeds and wind spectra but also provides an accurate time-varying coherence function. The results also manifest that if three anemometers can be equipped at appropriate locations of the bridge deck, a sufficiently accurate non-stationary wind field can be reconstructed.]]></description>
      <pubDate>Fri, 15 May 2026 15:44:12 GMT</pubDate>
      <guid>https://trid.trb.org/View/2670154</guid>
    </item>
    <item>
      <title>Comparative study on train-track flow field and tail car lift in high-speed EMS maglev trains with multiple marshalling lengths</title>
      <link>https://trid.trb.org/View/2670134</link>
      <description><![CDATA[This study investigates the critical effect of train marshalling length on the aerodynamic stability of a 600 km/h maglev train, focusing on the nonlinear behavior of tail car lift and the underlying flow mechanisms. Using the Improved Delayed Detached Eddy Simulation (IDDES) method validated by wind tunnel tests, we simulated configurations of 2 to 5 cars. The key finding is a nonlinear variation in the tail car's lift coefficient, which initially increases by up to 5% before decreasing by up to 22%, with a turning point at three cars. This phenomenon is driven by a fundamental shift in underbody flow: from axial discharge in short formations to lateral escape in long ones. The lateral flow generates large-scale vortices within the suspension gap; these vortices expand and propagate upstream with increasing train length, ultimately blocking the underbody airflow and severely deteriorating the gap environment. The results provide new insights into flow physics and direct implications for the aerodynamic design and safe operation of high-speed maglev systems.]]></description>
      <pubDate>Fri, 15 May 2026 15:44:12 GMT</pubDate>
      <guid>https://trid.trb.org/View/2670134</guid>
    </item>
    <item>
      <title>A study on buffeting response analysis: Rational functions vs Stack state-space method</title>
      <link>https://trid.trb.org/View/2673097</link>
      <description><![CDATA[Buffeting response theory and analysis has been the basis for the derivation of design wind loads on long-span bridges for over six decades. The wind pressure fluctuations about bridge sections, e.g., over a deck or tower, result in complex load patterns and these flexible elements start to respond and move. This movement is known to modify the loads – a phenomenon referred to as self-excited forcing. These self-excited forces need to be described theoretically to accurately predict the dynamic response of a bridge to turbulent wind. The practical application to bridge design consists primarily of two fundamental explanations: a) the quasi-static, and b) the unsteady aerodynamic theories. Based on the adopted theoretical model an appropriate response solving method is applied. Aerodynamic derivatives are commonly used for estimation of self-excited loads. A shortcoming in the application of derivatives is their mixed time and frequency dependence which poses problems to both wind stability and response analysis for multi degree-of-freedom (DOF) complex structures such as long-span bridges. Even though the structural modes of vibration are inherently uncoupled, aerodynamic coupling may be present between modes. An alternative to the mixed frequency-time domain formulation is the use of rational functions. These are frequency independent functions that are continuous in time offering attractive possibilities for improved response predictions. In this study, the response predictions based on the rational function approach are compared to those obtained via aerodynamic derivatives applying the proposed Stack State-Space response analysis method. It is observed that when the fit to the aerodynamic derivatives is close then the two methods yield similar results; however, differences in response are observed as the quality of fit worsens.]]></description>
      <pubDate>Fri, 15 May 2026 15:44:12 GMT</pubDate>
      <guid>https://trid.trb.org/View/2673097</guid>
    </item>
    <item>
      <title>Long-term wind-induced response of suspension bridges including static response, flutter stability, and parametric effects of turbulence</title>
      <link>https://trid.trb.org/View/2667103</link>
      <description><![CDATA[Parametric effects induced by atmospheric turbulence have emerged as an important factor influencing the aeroelastic behavior and extreme response of long-span suspension bridges. Originating from angle-of-attack fluctuations due to large-scale turbulence, these effects can significantly modify aerodynamic damping and stiffness, particularly for streamlined bridge decks. Long-term analysis, mainly adopted in the field of offshore structures, overcomes some limitations of classical Davenport theory-based approaches for calculating the dynamic response to turbulent wind of flexible structures, such as long-span suspension bridges. Among other aspects, it accounts for the influence of the statistical variability in turbulence parameters on the structural response, which is expected to impact on the actual role played by parametric effects of turbulence. However, accounting for these effects typically requires time-domain simulations, leading to prohibitive computational costs. This study introduces an efficient frequency-domain framework that incorporates the most significant parametric effect of turbulence (the so called “average parametric effect”) into the long-term evaluation of extreme response. The proposed formulation also includes static response and flutter instability, two aspects usually overlooked in previous contributions. The methodology is applied to the Halsafjorden Bridge, a planned 2000-m span suspension bridge in Norway. Three different wind scenarios, in terms of turbulence intensity and mean wind speed, are also considered. Long-term extremes are close to the results of the classical short-term approach if the mean wind speed is the only environmental random variable. In contrast, non-negligibly larger long-term responses are obtained if the randomness in turbulence intensity is also considered. Moreover, results reveal that the parametric effects of turbulence can significantly increase the long-term extreme response, particularly in torsion, where turbulence-induced damping reductions may lead to response increments of up to 41% for a return period of 100 years. Their impact is greater than in classical short-term analyses, where the average parametric effect leads to an increase in the torsional response of about 33%. This behavior is even more pronounced for higher return periods. These findings highlight that the combined influence of parametric effects of turbulence and randomness in the environmental parameters (e.g., turbulence intensity) can properly be assessed only within a long-term analysis.]]></description>
      <pubDate>Thu, 07 May 2026 09:20:59 GMT</pubDate>
      <guid>https://trid.trb.org/View/2667103</guid>
    </item>
    <item>
      <title>Full-scale assessment of vehicle wind loading on the Great Belt East Bridge</title>
      <link>https://trid.trb.org/View/2667081</link>
      <description><![CDATA[The present paper reports on measurements of wind pressures made on a full-scale box trailer travelling across the Great Belt East Bridge. The purpose of the measurements was to assess the efficiency of local wind screens designed to alleviate the increased wind loading and decrease the recorded risk of wind induced overturning in the vicinity of the bridge towers and the anchor blocks. Several conclusions are drawn from the present investigation. The wind loading on the trailer is a function of the depth of the bridge section on which it is travelling. It increases by 75% when running on the 4.4 m deep girder of the suspension bridge compared to the 7 m deep girder of the approach bridges. The full-scale turbulent wind loading on the stationary trailer is found to display significantly higher low frequency components than observed in the model scale wind tunnel tests. This is not surprising in view of the wind tunnel turbulence modelling technique but should be remembered when assessing wind tunnel test results. Finally, it is concluded that the wind screens serve to smoothen the wind pressure on the trailer at the towers and anchor blocks.]]></description>
      <pubDate>Thu, 07 May 2026 09:20:59 GMT</pubDate>
      <guid>https://trid.trb.org/View/2667081</guid>
    </item>
    <item>
      <title>Neural network-based eddy viscosity prediction for bluff body vehicle wake flow at high Reynolds number</title>
      <link>https://trid.trb.org/View/2666938</link>
      <description><![CDATA[Improvements in RANS model accuracy and computational efficiency can substantially impact automotive aerodynamic optimization. In this study, a deep neural network–based turbulence model is developed to predict the flow around bluff body vehicles. Using an Ahmed body dataset generated by the SST k−ω model, a mapping between flow-field physical quantities and eddy viscosity is constructed. The input features are extended to capture three-dimensional flow characteristics, with a random forest algorithm employed for feature selection. Analysis reveals that the orthogonality between velocity and its gradient, previously less significant in two-dimensional flows, becomes critical for predicting three-dimensional Ahmed body turbulence. The proposed model fully replaces the conventional SST k−ω model and is coupled with the CFD solver. Results show that it accurately reproduces the velocity and pressure fields, closely matching baseline RANS predictions. The predicted drag coefficient deviates by less than 6% from experimental measurements. For off-training conditions at the different yaw angle, the model exhibits slight underprediction in the wake core region and minor discrepancies in capturing upper vortices. Moreover, the model achieves a 30% reduction in computational time, demonstrating its potential for efficient, high-fidelity aerodynamic simulations in industrial applications.]]></description>
      <pubDate>Wed, 29 Apr 2026 16:34:58 GMT</pubDate>
      <guid>https://trid.trb.org/View/2666938</guid>
    </item>
    <item>
      <title>Adaptive modal pushover analysis for efficient buffeting performance evaluation of long-span bridge decks</title>
      <link>https://trid.trb.org/View/2659391</link>
      <description><![CDATA[The long-span bridge decks are susceptible to wind-induced vibrations due to their high flexibility and low damping. Considering the potential material savings by allowing the nonlinearity in structural elements under strong winds, the study is motivated by the recent performance-based wind design methodology to evaluate and understand the inelastic behaviors of long-span bridge decks at multiple buffeting performance levels. While the nonlinear time history analysis can offer very detailed wind structural response information, the required volume of computations is significant due to the long duration of windstorms. Hence, the static nonlinear analyses at multi-level wind hazards (i.e., wind buffeting pushover analysis) are explored in this study to efficiently provide adequate information on wind demands of the bridge deck and its components. To this end, the conventional equivalent static wind loads (ESWLs) for the linear elastic buffeting analysis is extended into the nonlinear inelastic regime, with the consideration of higher structural modes, inelastic behaviors, and multi-location responses. Inspired by the modal pushover analysis procedure for seismic demand evaluation and load-response-correlation method for wind load distribution estimation, the peak displacements at multiple bridge deck locations considering contributions from multiple modes and their coupling effects are first obtained using the pseudo-excitation method, and then the ESWLs are acquired using the displacement influence line. Furthermore, the structural characteristics (e.g., modal properties and displacement influence lines) are updated at each step of the pushover analysis to consider the effects of bridge deck inelastic behaviors on the ESWLs. A long-span truss bridge deck is employed as the case study to demonstrate the high accuracy and efficiency of the developed adaptive modal pushover analysis (AMPA) procedure for buffeting performance evaluation. Based on the inelastic behavior evolution of bridge deck elements with the increase of wind intensity, four buffeting performance levels are identified on the capacity curve. Finally, the sensitivity analysis is conducted to examine the contributions of multiple-mode and inelastic considerations to the wind demands estimated with AMPA.]]></description>
      <pubDate>Tue, 21 Apr 2026 08:28:11 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659391</guid>
    </item>
    <item>
      <title>Extrema estimation of vehicle-bridge interaction responses under crosswind via probabilistic decoupling</title>
      <link>https://trid.trb.org/View/2659427</link>
      <description><![CDATA[This study proposes a novel probabilistic method for estimating the extrema of the vehicle-bridge system subjected to crosswind and track irregularities, with a focus on key safety indexes such as wheel unloading ratio and derailment coefficient. The low-frequency components of crosswind induce significant variability in the vehicle's response during its short passage over the bridge. To address this, a probabilistic decoupling framework is introduced, in which the total system response is approximated as the sum of an intermediate response, due to crosswind alone, and a stationary Gaussian increment process, arising from track irregularities and coupling effects. A moving average filtering technique is employed to achieve this decomposition. An extreme value estimation formula is developed based on decomposition and upcrossing rate theory, and it is validated through numerical simulations. The method shows good accuracy and low bias using only tens of samples, outperforming conventional approaches. Finally, the proposed method is applied to evaluate the safe operating speed of high-speed vehicles under a mean wind velocity of 30 m/s. Results indicate that the wheel unloading ratio is the dominant safety index, and safe operation can be ensured at speeds up to 200 km/h.]]></description>
      <pubDate>Tue, 21 Apr 2026 08:28:11 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659427</guid>
    </item>
    <item>
      <title>Investigation of train-embankment relative motion modes on aerodynamic performance of a high-speed train under crosswind</title>
      <link>https://trid.trb.org/View/2659426</link>
      <description><![CDATA[The high-speed train (HST) running on embankments faces safety risks from intense crosswinds, potentially leading to derailments. Accurate simulation of the relative motion modes under crosswind is crucial to understanding aerodynamic load variations. This study investigates the aerodynamic characteristics of an HST on an embankment using two numerical methods: the static synthesis method (SSM) and the dynamic decomposition method (DDM). The improved delayed detached eddy simulation (IDDES) approach was used to analyse the flow field around the train, comparing aerodynamic loads, pressure distributions, and flow characteristics under two motion modes. Results show that varying velocity-inlet boundaries significantly impact flow characteristics around the embankment during strong crosswinds. In contrast, the SSM and the DDM effectively reduce the train's side force coefficient (Cy), lift force coefficient (Cz), and overturning moment (Cmx) by 16.9 %, 12.1 %, and 18.5 %, respectively. These findings provide important data to support the formulation of operational standards for high-speed trains (HSTs) running on embankments.]]></description>
      <pubDate>Tue, 21 Apr 2026 08:28:11 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659426</guid>
    </item>
    <item>
      <title>A wind-tunnel study of the effect of sheared wind profiles on the aerodynamic drag of passenger vehicle models</title>
      <link>https://trid.trb.org/View/2659424</link>
      <description><![CDATA[The effect of shear and skewness of the apparent wind approaching a passenger vehicle, due to variations of the terrestrial wind speed with height, was investigated. Based on terrestrial wind profiles measured adjacent to a test track, two highly sheared and skewed apparent-wind profiles were simulated at 15% scale in a wind tunnel to determine their effect on the aerodynamic drag and surface pressures of passenger vehicle models. The results show that the common track-test procedure of measuring the reference wind speed and angle at vehicle half-height, without regard for vertical variations, can underestimate the drag coefficient by 1% to 12% in windy conditions. Using a reference wind speed based on the average wind speed, or the average squared wind speed, over the height of the vehicle improved the prediction, although the discrepancy was still up to 7%. These averaged reference wind speeds, as well as the equivalent apparent-wind angle at which a uniform profile would produce the same drag coefficient as a sheared and twisted profile, were lower than the apparent-wind speed and angle at vehicle half-height for both simulated profiles. Importantly, they also occurred below vehicle half-height in the simulated profiles. The drag coefficient in the sheared and twisted apparent winds was lower than for uniform crosswinds at the same half-height yaw angle. This highlights the strong influence of flow conditions around the lower half of the vehicle on the aerodynamic drag coefficient, which was corroborated by surface pressure data. The implications for track testing are that, in the absence of apparent-wind profile measurements, it would be more appropriate to measure the reference apparent-wind speed and angle below rather than at vehicle half-height, in agreement with the recommendations of other researchers. The results of this study also have implications for wind-averaged drag computations.]]></description>
      <pubDate>Tue, 21 Apr 2026 08:28:11 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659424</guid>
    </item>
    <item>
      <title>CFD analysis of exhaust flow for reducing soot stains on railcar body surfaces</title>
      <link>https://trid.trb.org/View/2659423</link>
      <description><![CDATA[Diesel railcars are widely used in rail transport, particularly in rural areas, because of their ability to operate without overhead power lines. However, the exhaust gas emitted by diesel railcars can cause soot stains on the car body surface, which requires regular cleaning. In this study, computational fluid dynamics (CFD) simulations were conducted to investigate the effects of roof equipment and exhaust pipe configurations on the exhaust flow around a car body. Unsteady flow analysis was performed using delayed detached eddy simulation. The exhaust flow from the exhaust pipe was simulated using a non-isothermal flow based on the Boussinesq approximation. The velocity profiles obtained by CFD were validated against wind tunnel test results. The CFD results showed that the exhaust gas emitted into a cavity consisting of roof equipment caused soot staining on the car body surface. This study proposes an appropriate location for the exhaust outlet, in which the flow velocity normalised to the train speed was higher than 0.7 to reduce soot stains on the surface.]]></description>
      <pubDate>Tue, 21 Apr 2026 08:28:11 GMT</pubDate>
      <guid>https://trid.trb.org/View/2659423</guid>
    </item>
    <item>
      <title>Aerodynamic performance and operational safety of high-speed trains in crosswind: visualization and analysis of mapping knowledge domain</title>
      <link>https://trid.trb.org/View/2656063</link>
      <description><![CDATA[High-speed rail systems deliver substantial capacity, efficiency, and reliability, yet their susceptibility to crosswind effects increases with higher operating speeds. Employing a mapping knowledge domain (MKD) approach, this study provides a comprehensive analysis of the development trends in research on aerodynamic performance and operational safety of trains in crosswind environments, drawing on papers published from 1981 through 2025 in the Science Citation Index Expanded (SCIE) and Social Science Citation Index (SSCI). Knowledge mapping tools VOSviewer and Sci2 Tool are utilized, with key findings as follows: (1) Publication output exhibited gradual growth from 1981 to 2009, followed by a marked increase thereafter and China has emerged as the principal contributor to research on this field, particularly in engineering applications; (2) Multidimensional network analyses of authorship, institutional, and international collaborations reveal cooperative links among leading universities and research teams worldwide; (3) Document co-citation and keyword co-occurrence analyses have been employed to delineate subfields and synthesize frontier advancements within each; (4) Burst detection analysis reveals emerging research trends, such as “wind barrier” and “dynamic response”, which are likely to become key areas of focus in ensuring the stable and safe operation of trains under crosswind conditions in the future. This review provides scholars with a coherent, comprehensive framework and practical guidance for investigating train aerodynamic characteristics and ensuring operational safety.]]></description>
      <pubDate>Wed, 15 Apr 2026 08:31:47 GMT</pubDate>
      <guid>https://trid.trb.org/View/2656063</guid>
    </item>
    <item>
      <title>Aerodynamic characteristics and POD analysis of the long-span bridge under non-uniform flows</title>
      <link>https://trid.trb.org/View/2655897</link>
      <description><![CDATA[The present study investigates the three-dimensional aerodynamic characteristics under non-uniform inflow through wind tunnel tests in comparison with uniform flow. A 4-m-long test model was employed, with ten cross-sectional pressure-measuring planes. The study examines the spanwise distribution of surface pressure and aerodynamic force coefficients, as well as their frequency characteristics. The relationship between wake frequencies, aerodynamic force frequencies, and trailing-edge surface pressure frequencies was analyzed. Proper Orthogonal Decomposition (POD) was applied to the time-frequency analysis results of the wake field. The absolute values of POD modes represent the extracted wake frequency components. Both POD and time-frequency analysis quantitatively characterize the spatial non-uniformity, spectral complexity, and temporal variability of wake frequencies under non-uniform flow. The results indicate that non-uniform wind velocities have a small impact on the mean pressure coefficients but a significant effect on the fluctuating pressure coefficients and aerodynamic force fluctuation coefficients. Larger wind velocity differences in non-uniform profiles lead to more pronounced spanwise non-uniformity in fluctuating pressure coefficients, mean lift coefficients, and POD modal energy distributions. The spanwise distribution of surface pressure frequencies at the trailing edge, high-frequency regions of aerodynamic forces, and wake frequencies is aligned with the trend of the inflow velocity profile.]]></description>
      <pubDate>Wed, 15 Apr 2026 08:31:47 GMT</pubDate>
      <guid>https://trid.trb.org/View/2655897</guid>
    </item>
    <item>
      <title>Dynamic characteristics of overhead electrification catenary support structure in high-speed railway tunnel under train slipstream: A FSI simulation study</title>
      <link>https://trid.trb.org/View/2655812</link>
      <description><![CDATA[When the high-speed train passes through the tunnel, its rapid movement causes severe air disturbance, leading to complex and intense train slipstream effects. Under these slipstream conditions, the Overhead Electrification Catenary Support Structure (OESS) inside the tunnel inevitably interacts with the transient airflow through fluid-structure interaction, consequently inducing complex vibrational responses. This study investigates the dynamic response characteristics of OESS in high-speed railway tunnels under train-induced slipstream effects using a three-dimensional fluid-structure interaction model. The results demonstrate that the longitudinal aerodynamic loads dominate the structural response, inducing significantly higher displacements and accelerations compared to the lateral and vertical directions. Notably, it is found that shorter train formations generate more critical aerodynamic excitation than longer formations, producing higher dynamic responses and load magnitudes. Quantitative analysis reveals distinct power-law relationships between train speed and OESS response parameters, while tunnel cross-sectional area shows linear correlations. Aerodynamic loads distribute non-uniformly across OESS components, with the Mast Pole experiencing the highest load intensity and the Steady Arm the lowest. Mechanistic insight from flow field analysis demonstrates that the enhanced responses under shorter formations originate from substantially increased local wind speeds (by over 10 %), elevated turbulence intensity, and more pronounced vortex structures. These findings provide critical insights for the aerodynamic safety design and fatigue assessment of OESS in high-speed railway tunnels.]]></description>
      <pubDate>Mon, 30 Mar 2026 17:15:44 GMT</pubDate>
      <guid>https://trid.trb.org/View/2655812</guid>
    </item>
    <item>
      <title>An explainable deep ensemble model for probabilistic prediction of typhoon effects on a long-span bridge</title>
      <link>https://trid.trb.org/View/2652419</link>
      <description><![CDATA[Long-span bridges usually suffer severe vibrations under extreme wind events (such as typhoons), potentially leading to engineering failures and traffic accidents. Data-driven approaches facilitate the mitigation of risks through timely and accurate prediction of typhoon effects. Deep learning (DL) algorithms, including the convolutional neural network (CNN), long short-term memory (LSTM), and their combined models, have been extensively applied in various fields. Despite the superior predictive performance of CNN-LSTM in time series, it fails to provide probabilistic estimates to quantify uncertainty and lacks adequate interpretability. In this work, a CNN-bidirectional LSTM-based explainable deep ensemble (CNN-BiLSTM-EDE) model is proposed for the probabilistic prediction of typhoon effects on long-span bridges. Specifically, CNN and BiLSTM are integrated to enhance the capability of processing spatiotemporal typhoon characteristics. A deep ensemble scheme is then adopted to modify the CNN-BiLSTM architecture, enabling dynamic response estimation within a probabilistic framework. The final prediction is obtained by averaging the results through ensemble learning. Shapley additive explanation (SHAP) is introduced to reveal the marginal contributions and substantive impacts of feature variables on the model predictions. Decade-long typhoon datasets of a kilometer-scale long-span bridge are utilized to validate the proposed model. The results indicate that CNN-BiLSTM-EDE provides reliable response predictions while quantifying uncertainty by offering corresponding conditional distribution. According to the SHAP visualization results, mean wind speed and wind direction angle are identified as the most influential factors in predicting typhoon effects. Compared with four probabilistic benchmark models, CNN-BiLSTM-EDE demonstrates superior prediction accuracy and uncertainty quantification performance.]]></description>
      <pubDate>Thu, 26 Mar 2026 16:59:30 GMT</pubDate>
      <guid>https://trid.trb.org/View/2652419</guid>
    </item>
  </channel>
</rss>