Vortex-Induced Drag Forecast for Cylinder in Non-uniform Inflow
Jiashun Guan (关家顺), Haoyang Hu (胡皓阳), Tianfang Hao (郝天放), Huimin Wang (王慧敏), Yunxiao Ren (任耘霄)*, Dixia Fan (范迪夏) · 2025
Read Vortex-Induced Drag Forecast for Cylinder in Non-uniform Inflow by Jiashun Guan (关家顺), Haoyang Hu (胡皓阳), Tianfang Hao (郝天放), Huimin Wang (王慧敏), Yunxiao Ren (任耘霄)*, Dixia Fan (范迪夏) (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-2506-21075.txt.
Introduction
Jiashun Guan and co-authors forecast vortex drag on a cylinder in uneven flow in a 2025 letter. Offshore gear faces drag that jumps as vortices shed, and streaky inflow scrambles the pattern. Past pressure based calls falter there. Their tuned neural net adds upstream speed reads as a flow gauge.
Direct simulations at a set Reynolds number trained and checked the net. Sensor spot and speed part choices were tuned step by step. The best layout foresaw sharp drag swings one time step ahead with a fair score. Sparse but well placed sensors did the work, tied to flow split points.
Catalog record
| CREATOR | Jiashun Guan (关家顺), Haoyang Hu (胡皓阳), Tianfang Hao (郝天放), Huimin Wang (王慧敏), Yunxiao Ren (任耘霄)*, Dixia Fan (范迪夏) |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 21 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-21075
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Vortex-Induced Drag Forecast for Cylinder in Non-uniform Inflow Preprint: AIP/123-QED Jiashun Guan (关家顺), Haoyang Hu (胡皓阳), Tianfang Hao (郝天放), Huimin Wang (王慧敏), Yunxiao Ren (任耘霄)*, Dixia Fan (范迪夏) Abstract In this letter, a physics-based data-driven strategy is developed to predict vortex-induced drag on a circular cylinder under non-uniform inflow conditions – a prevalent issue for engineering applications at moderate Reynolds numbers. Traditional pressure-signal-based models exhibit limitations due to complex vortex dynamics coupled with non-uniform inflow. To address this issue, a modified fully connected neural network (FCNN) architecture is established that integrates upstream velocity measurements (serving as an inflow calibration) with pressure-signal-based inputs to enhance predictive capability ( R 2 ∼ 0 → 0.75 R^{2}\sim 0\rightarrow 0.75 ). Direct numerical simulations (DNS) at Reynolds number R e = 4000 Re=4000 are implemented for model training and validation. Iterative optimizations are conducted to derive optimized input configurations of pressure sensor placements and velocity components at upstream locations. The optimized model achieves an R 2 R^{2} score of 0.75 in forecasting high-amplitude drag coefficient fluctuations ( C d = 0.2 CLOSE (C_{d}=0.2 – OPEN 1.2 ) 1.2) within a future time window of one time unit. An exponential scaling between model performance and optimized pressure signal inputs is observed, and the predictive capability of sparsely distributed but optimized sensors is interpreted by the scaling. The optimized sensor placements correspond to the physical mechanism that the flow separation dynamics play a governing role in vortex-induced drag generation. This work advances machine learning applications in fluid-structure interaction systems, offering a scalable strategy for forecasting statistics in turbulent flows under real-world engineering conditions. The flow around cylindrical structures has attracted sustained research attention for over a hundred years Zdravkovich 1997 ; Von Kármán 1911 ; Konstantinidis and Liang 2011 ; Norberg and Sunden 1987 ; Williamson 1996 ; Lin et al. 2021 ; Kravchenko and Moin 2000 , motivated by their fundamental flow configuration and practical implementations encompassing ocean engineering systems, wind engineering applications, building aerodynamics, and so on. In real-world engineering applications, the non-uniform inflow conditions always introduce turbulence effects Bearman and Morel 1983 , exhibiting more complex characteristics relative to wake dynamics of cylinder flows Williamson 1996 ; Bhattacharyya et al. 2022 , e.g. the shrunk recirculation bubbles of the mean flow and high-amplitude fluctuations in drag (lift) coefficient C_{d} ( C_{l} ) Song et al. 2022 . Figure 1: (a) The spatial distribution of spectral elements and the Z -component of vorticity in the X-Y cross-sectional plane ( Z=1.5 ). The insert is zoomed view of the red box, and red points indicates uniformly distributed locations of monitored points. The asterisk in the insert indicates the first monitored point ( pt=1 ), with label number increasing clockwise around the circle. (b) Visualization of turbulent structures through vortex criterion Q=1.5~U^{2}/D^{2} , colored by the Z -component of vorticity. To date, measuring Bhattacharyya et al. 2023a ; Guan et al. 2021 ; Khan et al. 2023 and forecasting Barthel and Sapsis 2023 ; Lee and You 2019 dynamic evolution processes in fluid-structure interaction systems present technical challenges, while remaining critically important for preventing structure damage induced by extreme events Sapsis 2021 and frequency lock-in Bearman and Morel 1983 ; Lin et al. 2021 . By leveraging machine learning and data-driven methods Fan et al. 2020 ; Brenner, Eldredge, and Freund 2019 ; Brunton, Noack, and Koumoutsakos 2020 , progresses are achieved in areas such as statistics prediction, flow field reconstruction, flow control, and inverse problem resolution. Barthel and Sapsis Barthel and Sapsis 2023 proposed a data-driven method for predicting airfoil extreme events using surface pressure signals processed by wavelet transformation, enabling accurate forecasting with a simple feed-forward neural network and minimal sensor deployment. Kim and Sapsis Kim and Sapsis 2024 achieved real-time lift coefficient prediction for a dynamically stalling airfoil by employing discrete wavelet transforms to process surface pressure signals. Figure 2: Under non-uniform inflow conditions, the following measurements are presented: drag coefficient C_{d} , pressures p at pt=4 (blue) and pt=21 (red), streamwise velocity u at (-1,0,0) , and continuous wavelet transform (CWT) processed C_{d} signal. The contour in the bottom panel represents amplitudes of the frequencies f , linearly spanning the range [0,0.1] . While the aforementioned works have demonstrated success in airfoil lift prediction through wavelet-processed pressure signals and neural networks, the pure pressure-signal-based model faces inherent limitations when applied to cylinder flow problems – particularly for drag coefficient prediction under non-uniform inflow conditions that commonly characterize cylinder flows in real-world engineering applications at moderate Reynolds number regimes. In such cases, the complex vortex shedding dynamics, three-dimensional wake instabilities, and spatially heterogeneous pressure distributions induced by non-uniform inflows weaken the direct correlation between surface pressure signatures and drag force statistics. By incorporating physical interpretations to circumvent conventional brute-force testing approaches, this study therefore establishes a modified modeling framework specifically designed to be capable of these challenges. Direct numerical simulations (DNS) of flow over a circular cylinder are implemented at Reynolds number Re=UD/\nu=4000 , where U , D and \nu are the flow rates, cylinder diameter and kinematic viscosity, respectively. The flow is governed by the incompressible Navier-Stokes equations: \frac{\partial\mathbf{u}}{\partial t}+(\mathbf{u}\cdot\nabla)\mathbf{u}=-\nabla p+\frac{1}{Re}\nabla^{2}\mathbf{u} (1a) \nabla\cdot\mathbf{u}=0 (1b) where \mathbf{u} is the vector of velocities (u,v,w) and p is the pressure field. The equations are nondimensionalized by setting the flow rates U and the cylinder diameter D as the characteristic velocity and length scales,respectively, and solved by the open source spectral element code Nek5000 Paul, Fischer, and Kerkemeier 2008 . Periodic boundaries are applied in streamwise, spanwise, and cylinder-paralleled directions, respectively, leading to a non-uniform inflow of the cylinder. Spatial discretizations utilize 17020 spectral elements in 7^{th} order accuracy. The periodic cylinder is in 3~D length and discretized into 5 elements. The streamwise-spanwise sectional view of spectral elements are shown in Fig. 1 a. The time stepping is set to 0.0005~D/U , resulting in a CFL number of approximately 0.5 to ensure numerical stability. The simulation spans a total dimensionless time of 3000~D/U , comprising an initial transient phase ( 300~D/U ) for flow development and a subsequent statistically stationary period ( 2700~D/U ) under non-uniform inflow conditions for turbulence statistics acquisition. Velocities and pressures at 32 points along the cylinder and three upstream points located at (-2,0,0) , (-1,0,0) , and (-0.7,0,0) (shown in Fig. 1 a), are dumped every 20 time steps ( \Delta t=0.01~D/U ). The numerical setup is validated by comparing time-averaged ( 1000D/U ) pressure distributions under uniform inflow conditions against previous experimental results Norberg 1994 , demonstrating good agreement within 3.4\% deviation across the monitored point locations. Figure 3: Architecture diagram of the fully connected neural network (FCNN), where R^{i} represents the length of each fully connected layer. Numerical results in Fig. 2 are partitioned into training and testing sets ( 2400~D/U and 300~D/U , respectively) for subsequent C_{d} forecasts using a fully connected neural network (FCNN), which employs the architecture: input-FC24-FC28-FC24-FC16-FC8-output (Fig. 3 ). The output layer with fixed length 1 is the predicted drag coefficient C_{d} at a future time window of \Delta t_{p}=1~D/U (unless otherwise specified in this study), while the hidden layer maintains a fixed architecture. Note that the input layer exhibits variable dimensions, and