Zhao Wu*, Jin Lee, Charles Meneveau, Tamer Zaki, Department of Mechanical Engineering, Johns Hopkins University
An unsupervised machine-learning algorithm, the self-organizing map (SOM), is used to identify the turbulent boundary layer (TBL) and non-TBL regions in bypass transition. The data employed for the analysis are from an archived direct simulation publicly available in the Johns Hopkins Turbulence Databases (JHTDB, http://turbulence.pha.jhu.edu), stored using the new FileDB system. The data points in the entire flow domain are automatically classified into TBL and non-TBL regions by the SOM, based on their standardized velocity, velocity fluctuations, velocity gradients and their spatial locations. Thus the SOM identifies the turbulent-boundary-layer interface (TBLI) without the usual need for choosing thresholds on e.g. vorticity or velocity fluctuations. The TBLI is found to be a hyperplane in the input space. The SOM distinguishes the streaks in the laminar region and the weak free-stream turbulence from TBL region. Results from our approach are shown to be consistent with threshold-based methods in the special cases when those are applicable.
The authors acknowledge fundings from Office of Naval Research and National Science Foundation. Computations were made possible by Maryland Advanced Research Computing Center and Extreme Science and Engineering Discovery Environment.