Surface Line Integral Convolution-Based Vortex Detection Using Computer Vision

Abstract Vortex cores in fluid mechanics are easy to visualize, yet difficult detect numerically. Precise knowledge of these allows dynamics researchers study complex flow structures and allow for a better understanding the turbulence transition process development evolution instabilities, name but few relevant areas. Various approaches such as Q, delta, swirling strength criterion have been proposed visualize vortical flows, can be used vortex core locations. Using methods result spuriously detected which balanced by cutoff filter, making lack robustness. To overcome this shortcoming, we propose new approach using convolutional neural networks directly from streamline plots, line integral convolution method. We show that our computer vision-based is able reduce number false positives negatives while removing need cutoff. validate Taylor–Green problem generate input images network. with an increasing training, monotonically negatives. then apply trained network different where vortices still reliably detected. Thus, presents robust reliable detection applicable wide range scenarios.