Detecting the large-scale wall-attached structural inclination angles by a machine learning perspective in turbulent boundary layer.

With the recent advances in machine learning, strategies based on data can be used to augment wall modeling in the turbulent boundary layer. Combined with the attached eddy hypothesis, the present work applies extreme gradient boosting (XGBoost) to predict the large-scale wall-attached structures at a range of wall-normal locations based on a near-wall reference position ( z R + ≈ 4) spanning a Reynolds-number range R e τ ∼ O (10 3) − O (10 5). The input and output signals are selected as the large-scale structures; here, the input signals are set as in the fixed near-wall reference position by a series of streamwise velocity ({ X − N , ... , X − 1 , X 0 , X 1 , ... , X N }), and the output signal Y 0 is set directly above X 0 . Within each dataset, the large-scale wall-attached structures are identified from the prediction modeled by XGBoost between the turbulence in the upper region and at the near-wall reference position, resulting in a successful prediction of the large-scale structures inclination angles. Along the wall-normal offset Δ z and streamwise offset L x (distance between X i and X 0 ), the slope of the feature importance (represented by contour levels) is exactly equal to the degree of inclination of large-scale structures, indicating the turbulent inner and outer connection inferred by the machine learning input and output interactions perspective. This study shows that there is a great opportunity in machine learning for wall-bounded turbulence modeling by connecting the flow interactions between near-wall and outer regions. [ABSTRACT FROM AUTHOR]