Estimation of Flow Features in the Wake of a Circular Cylinder Using Artificial Neural Network


Sahin B., CANPOLAT Ç., BİLGİLİ M.

Arabian Journal for Science and Engineering, 2024 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Publication Date: 2024
  • Doi Number: 10.1007/s13369-024-09763-3
  • Journal Name: Arabian Journal for Science and Engineering
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Metadex, Pollution Abstracts, zbMATH, Civil Engineering Abstracts
  • Keywords: Artificial neural network (ANN), Circular cylinder, Particle image velocimetry (PIV), Streamwise velocity, Transverse velocity, Vorticity
  • Çukurova University Affiliated: Yes

Abstract

Accurate estimation of the flow features passing around a cylinder’s surface plays a crucial role in scientific and industrial applications to save time and financial costs. In this study, artificial neural network (ANN) is used to estimate the flow features in the wake of a circular cylinder immersed in a free-stream flow for Reynolds numbers of Re = 1500, 4000, 7000, and 9600. In the developed ANN model, the time-averaged flow data measured by the particle image velocimetry (PIV) technique are taken into account. While the X-coordinate and Y-coordinate values in the wake regions of the circular cylinder are used as inputs of the ANN model, vorticity ω, streamwise velocity u, and transverse velocity v variables are defined as the output data. Modeling of flow characteristics with easily obtainable independent variables without the need for any flow and physical properties is considered. The predicted results show that the ANN model is well trained using the trainlm learning algorithm and has the best result from Re = 1500 with 0.9960 R, 1.3859 m/s MAE, and 1.7917 m/s RMSE for the streamwise velocity 〈u〉 prediction. The patterns of ANN and PIV data show good agreement. As a result, the outcomes of the present study prove that the current ANN algorithm can accurately predict the experimentally measured time-averaged velocity data by only using spatial coordinates as inputs, which provides invaluable information for fluid dynamics community.