Linear logistic regression with weight thresholding for flow regime classification of a stratified wake

A stratified wake has multiple flow regimes, and exhibits different behaviors in these regimes. We aim at automated classification of the weakly stratified and the strongly stratified turbulence flow regimes based on information available in a full Reynolds stress model. We first generate a direct numerical simulation (DNS) database covering a Reynolds number range from 10 000 to 50 000 and a Froude number range from 2 to 50. 80 to 100 independent realizations of temporal evolving simulations are computed to get converged flow statistics. These data are used for training and testing purposes. Next, we train a linear logistic regression classifier with weight thresholding for automated flow regime classification. The classifier is designed to identify the physics critical to the classification task. Trained against data at one case, the classifier is found to generalize well to other Reynolds and Froude number combinations. We come to two conclusions from the trained classifier results. First, the physics governing wake evolution is universal, and second, the classifier captures that physics. The novelty of this work lies in the new DNS database and the development of a new machine learning tool.