Spatiotemporal wall pressure forecast of a rectangular cylinder with physics-aware DeepUFNet
Junle Liu, Chang Liu, Yanyu Ke, Wenliang Chen, Kihing Shum, K.T. Tse, Gang Hu · 2025
Read Spatiotemporal wall pressure forecast of a rectangular cylinder with physics-aware DeepUFNet by Junle Liu, Chang Liu, Yanyu Ke, Wenliang Chen, Kihing Shum, K.T. Tse, Gang Hu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-03183.txt.
Introduction
Junle Liu and colleagues forecast how wind pressure evolves over the surface of a rectangular cylinder. That value is central to structural work. Wall pressure governs the forces wind creates on bluff bodies and how structures reply. Past deep learning work predicted mean or fluctuating pressure coefficients. But it did so mostly for a single snapshot with full spatial facts, not as a forecast that evolves.
The team builds DeepUFNet, a deep learning model that joins a UNet structure with a Fourier neural network. Control of physical high-frequency loss is set in training to lift scores. Wind tunnel tests with high-frequency pressure scanning on two-dimensional rectangular cylinders gave the training and test database. The model forecasts wall pressure in space and time with accuracy for a cylinder of side ratio 1.5.
Catalog record
| CREATOR | Junle Liu, Chang Liu, Yanyu Ke, Wenliang Chen, Kihing Shum, K.T. Tse, Gang Hu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-03183
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Spatiotemporal wall pressure forecast of a rectangular cylinder with physics-aware DeepUFNet Preprint: APS/123-QED Junle Liu, Chang Liu, Yanyu Ke, Wenliang Chen, Kihing Shum, K.T. Tse, Gang Hu Abstract The wall pressure is of great importance in understanding the forces and structural responses induced by fluid. Recent works have investigated the potential of deep learning techniques in predicting mean pressure coefficients and fluctuating pressure coefficients, but most of existing deep learning frameworks are limited to predicting a single snapshot using full spatial information. To forecast spatiotemporal wall pressure of flow past a rectangular cylinder, this study develops a physics-aware DeepU-Fourier neural Network (DeepUFNet) deep learning model. DeepUFNet comprises the UNet structure and the Fourier neural network, with physical high-frequency loss control embedded in the model training stage to optimize model performance, where the parameter β \beta varies with the development of the training epoch. Wind tunnel testing is performed to collect wall pressures of a two-dimensional rectangular cylinder with a side ratio of 1.5 at an angle of attack of zero using high-frequency pressure scanning, thereby constructing a database for DeepUFNet training and testing. The DeepUFNet model is found to forecast spatiotemporal wall pressure information with high accuracy. The comparison between forecast results and experimental data presents agreement in statistical information, temporal pressure variation, power spectrum density, spatial distribution, and spatiotemporal correlation. It is also found that embedding a physical high-frequency loss control coefficient β \beta in the DeepUFNet model can significantly improve model performance in forecasting spatiotemporal wall pressure information, in particular, in forecasting high-order frequency fluctuation and wall pressure variance. Furthermore, the DeepUFNet extrapolation capability is tested with sparse spatial information input, and the model presents a satisfactory extrapolation ability. I Introduction Wall pressure for rectangular cylinders is of great importance to understand the fluid-induced forces of rectangular cylinders ( Yang et al. 2020 ) , structural responses ( Mills et al. 2002 ; Zhao et al. 2019 ) , civil structural safety, and building aerodynamics ( Hussain and Lee 1980 ; Zhang and Gu 2008 ) . Wind tunnel testing ( Knisely 1990 ; Okada and Ha 1992 ; Li et al. 2021 ) and numerical simulation ( Tan et al. 2004 ; Bruno et al. 2010 ; Li and Li 2017 ; Mashhadi et al. 2021 ; Chiarini et al. 2022a ; Chiarini et al. 2022b ; Li et al. 2024a ) are considered as convincing approaches to obtain wall pressure information on rectangular cylinders. Existing research has accumulated a large amount of work that utilizes the wind tunnel testing method to collect and investigate wall pressure information for rectangular cylinders ( Okajima 1982 ; Norberg 1993 ; Liu et al. 2024a ) as well as employing numerical tools to acquire and understand the rectangular cylinder wall pressure ( Okajima 1990 ; Shimada and Ishihara 2002 ; Sohankar 2008 ) . As the new emerging technique spreads, artificial intelligence (AI) has achieved remarkable progress in temporal or spatiotemporal prediction of aerodynamic characteristics, showing advantages in the fitting of high-dimensional features ( Han et al. 2018 ; Zhu et al. 2019 ; Brunton et al. 2020 ; Ben-Uri et al. 2025 ) and the extraction of characteristics from the data ( Lusch et al. 2018 ; Maulik et al. 2019 ; Glaws et al. 2020 ; Murata et al. 2020 ; Fukami et al. 2021a ) . For example, some researchers used AI techniques to predict the flow field or wind speed. Zhang et al. 2024 introduced a hybrid deep learning model that combines a convolutional neural network (CNN)-long short-term memory network and microscale meteorological models to predict wind speed. Their approach explicitly integrates physical priors with data-driven components, improving both accuracy and interpretability in complex urban environments. Bastos et al. 2021 harnessed the synergy between UNet and CNN to fuel spatiotemporal wind forecasting. In parallel, several studies explored graph-based and Transformer-based architectures for spatiotemporal wind prediction. Pan et al. 2022 proposed a spatiotemporal graph transformer network that effectively captured both local spatial and long-range temporal dependencies across multiple wind measurement stations. As for pressure information prediction, some works employ artificial intelligence for wall pressure prediction and forecasting. Liu et al. 2024b constructed two types of AI algorithms, graphic attention networks and dense neural networks, to make bidirectional predictions between the pressure of the rectangular cylinder wall and the wake flow. Their model evaluations on the numerical data showed that AI methods can predict both instantaneous wall pressure and time-averaged wall pressure based on the wake flow input. Tian et al. 2020 developed a deep neural network (DNN) to predict mean and peak wall pressure coefficients on rectangular cylinder-shaped buildings. Based on the experimental data collected, the trained DNN model can make satisfactory predictions on the mean pressure coefficients. Fernández-Cabán et al. 2018 built an artificial neural network (ANN) integrating backpropagation training to predict the mean, root mean square, and peak pressure coefficients on three geometrically scaled rectangular cylinders. Comprehensive experimental data were collected to train the model. Their evaluations presented that the ANN model was capable of predicting accurate mean, peak, and RMS pressure coefficients compared to the experimental data. However, existing AI-based studies have focused mainly on using full-dimensional data to predict the instantaneous wall pressure distribution ( Liu et al. 2024b ) , the mean pressure distribution ( Weng and Paal 2022 ; Liu et al. 2025 ) , the fluctuation pressure coefficient ( Hu et al. 2020 ) , and the maximum pressure coefficient ( Tian et al. 2020 ) . Few works have focused on spatiotemporal wall pressure forecast. In particular, there is an obvious absence of spatiotemporal wind tunnel wall pressure measurements forecast. This is partly due to the challenges posed by wind tunnel data, such as measurement sparsity, noise contamination, and non-stationary flow characteristics ( Li et al. 2025 ; Santos et al. 2023 ; Chen et al. 2021 ) , which make spatiotemporal and high-fidelity forecasting more difficult. Furthermore, in some practical scenarios and engineering cases, full-dimension wall pressure information is not always available or easily accessible, and there is a scarcity or noise in the measured data ( Li et al. 2025 ; Santos et al. 2023 ; Chen et al. 2021 ) , invalidating existing instantaneous wall pressure prediction models. To address these limitations, we developed a hybrid deep learning framework, the physics-aware Deep U-Fourier neural network model (DeepUFNet), for robust spatiotemporal experimental wall pressure forecast. DeepUFNet is developed based on the UNet architecture and the Fourier neural network, and embeds high-frequency loss control in the model training stage to optimize performance. The uniform flow past a rectangular cylinder under zero angle of attack is tested in the closed-loop wind tunnel, which generates the database for DeepUFNet model training and testing. As investigated from literature, when SR\in(1,2) , the standard deviation value of the lift coefficient is very large, indicating a large fluctuation of the wall pressure ( Sohankar 2008 ) . When SR=1.5 , the vortex generated from the leading corner is focused mainly on the regions around the upper and lower wall ( Shimada and Ishihara 2002 ) . The visualized contours of instantaneous vortices show that the vortex near the upper and lower walls interacted