Recognition and Classification of Vortical Flows using Artificial Neural Networks and Graftieaux’s Identification Criteria

One of the main challenges of studying unsteady fluid mechanics is the modeling and analyzing of the nonlinear wing-vortex interaction. The topology of the wake structure can significantly alter the aerodynamic outcomes of the maneuvering object. Identifying and tracking vortical structures are difficult because existing methods require computationally expensive algorithms and human input. The application of machine learning techniques in classifying these structures has the potential to streamline vortex identification. Instead of using computationally expensive algorithms requiring fluid dynamic analysis to define largescale geometric vortex properties, a neural network could learn from a predefined training dataset and identify vortical structures from images of a flowfield. The objective of this study is to develop a convolutional neural network model to recognize and classify vortical structures in an unsteady flow. The application of this network on a vorticity field yields a distribution of vortex probabilities at each location in the flowfield, which is then used to identify vortex centroids. Recording the identified vortex centroids across multiple time steps allows certain parameters such as vortex trajectory and velocity to be calculated for further analysis of the wake structure. The performance of this method is then compared to that of existing methods for its accuracy and computational efficiency.