Feature Identification in Complex Fluid Flows by Convolutional Neural Networks
Recent efforts have shown machine learning to be useful for the prediction of nonlinear fluid dynamics. Predictive accuracy is often a central motivation for employing neural networks, but the pattern recognition central to the network function is equally valuable for purposes of enhancing our dynamical insight into confounding dynamics. In this paper, convolutional neural networks (CNNs) were trained to recognize several qualitatively different subsonic buffet flows over a high-incidence airfoil, and a near-perfect accuracy was performed with only a small training dataset. The convolutional kernels and corresponding feature maps, developed by the model with no temporal information provided, identified large-scale coherent structures in agreement with those known to be associated with buffet flows. An approach named Gradient-weighted Class Activation Mapping (Grad-CAM) was then applied to the trained model to indicate the importance of these feature maps in classification. Sensitivity to hyperparameters including network architecture and convolutional kernel size was also explored, and results show that smaller kernels are better at coherent structure identification than are larger kernels. A long-short term memory CNN was subsequently used to demonstrate that with the inclusion of temporal information, the coherent structures remained qualitatively comparable to those of the conventional CNN. The coherent structures identified by these models enhance our dynamical understanding of subsonic buffet over high-incidence airfoils over a wide range of Reynolds numbers.