A machine learning-augmented aerodynamic database of rectangular cylinders

Abstract

Rectangular cylinders submerged in a fluid encounter intricate aerodynamic forces, and the forces significantly influence the stability and safety of structures possessing a rectangular cross section. Although aerodynamic characteristics of these cylinders have been extensively studied, a comprehensive database cataloging these characteristics remains absent. This study conducted a large number of wind tunnel pressure testings to establish an aerodynamic database for rectangular cylinders with 2470 distinct configurations, including turbulent intensities ranging from 1% to 20%, side ratios ranging from 0.6 to 5, and wind attack angles ranging from 0° to 90°. The accuracy of the database was validated by data from the literature and wind tunnel force measurement experiments. More importantly, machine learning models were developed and have substantially expanded the experimental data, resulting in a comprehensive, continuous aerodynamic database for rectangular cylinders. By evaluating the model performance and verifying its generalization capability, the accuracy of the machine learning-augmented database is proved. This database is anticipated to serve as a critical reference for academic research and a practical reference for engineering applications.