Machine learning-based prediction of crosswind vibrations of rectangular cylinders. Issue 211 (April 2021)

• Record Type: Journal Article
• Authors: Lin, Pengfei; Hu, Gang; Li, Chao; Li, Lixiao; Xiao, Yiqing; Tse, K.T.; Kwok, K.C.S.
• Is Part Of: Journal of wind engineering and industrial aerodynamics. Issue 211(2021)
• Publication Date: 2021-04
• Subjects: Machine learning -- Rectangular cylinder -- Crosswind vibration -- Galloping -- Vortex-induced vibration; Wind-pressure -- Periodicals; Buildings -- Aerodynamics -- Periodicals; Pression du vent -- Périodiques; Constructions -- Aérodynamique -- Périodiques; Buildings -- Aerodynamics; Wind-pressure; Periodicals
• DOI: 10.1016/j.jweia.2021.104549
• Languages: English
• ISSNs: 0167-6105

Abstract

Abstract: Due to the complexity of crosswind vibrations of rectangular cylinders, current research on crosswind vibrations of rectangular cylinders mainly relies on expensive wind tunnel tests and time-consuming numerical simulation techniques. In this study, in order to evaluate crosswind vibrations of rectangular cylinders, machine learning method was used to build an efficient and effective prediction model for supplementing the above two research tools. 5 machine learning models based on decision tree regression, k -nearest neighbor regression, random forest, gradient boosting regression tree (GBRT) and histogram gradient boosting regression tree algorithms were trained based on the existing high-quality and reliable wind tunnel test datasets of crosswind responses of rectangular cylinders. The hyper-parameters were optimized by using particle swarm optimization method. 4 types of crosswind vibration phenomena, including over-coupled, coupled, semi-coupled and decoupled, were predicted. It was found that the GBRT model is capable of predicting crosswind responses of rectangular cylinders at side ratios from 0.75 to 3 and Scruton numbers from 0 to 150 under wind flow with turbulence intensities from 0 to 16%. Evidently, GBRT model can be an effective and economical method to study crosswind vibrations of rectangular cylinders and hence supplement traditional wind tunnel tests and numerical simulation techniques. Highlights: A database of crosswind vibrations of rectangular cylinders was created. Five machine learning models were trained to predict crosswind responses. Gradient boosting regression tree was the optimal one among five models. The optimal model is able to predict crosswind responses accurately.