weights.pth.tar (704.63 MB)
Download file

Surface line integral convolution-based vortex detection using computer vision

Download (704.63 MB)
posted on 22.02.2022, 21:24 by Hazem Ashor Amran AbolhollHazem Ashor Amran Abolholl
Vortex cores in fluid mechanics are easy to visualise, yet difficult to detect numerically. Precise knowledge of these allow fluid dynamics researchers to study the underlying complex flow structures with greater precision and allow for a better understanding of the turbulence transition process and the development and evolution of flow instabilities, to name but a few relevant areas. Various approaches such as the Q, delta and swirling strength criterion have been proposed to visualise vortical flows and these approaches can be used as well to detect vortex core locations. Using these methods will detect spurious vortex cores and the number of false positives and negatives need to be balanced through a threshold criterion, making these methods lack robustness. To overcome this shortcoming, we propose a new approach using convolutional neural networks to detect flow structures directly from streamline plots, using the line integral convolution method. We show that our computer vision-based approach is able to reduce the number of false positives and negatives entirely while removing the need to calibrate user-defined parameters which are flow problem-specific. We validate our approach for the well-known Taylor-Green vortex problem where we extract line integral convolution-based streamline plots on the centre planes of the domain which are then used to train our convolutional neural network. We show that with an increasing number of images used for training, we are able to monotonically reduce the number of false positives and negatives. We then apply our trained network to a different flow problem and show that we are able to detect vortices irrespective of the flow case. Thus, our study presents a convolutional neural network approach that allows for reliable vortex core detection that only needs to be trained once but is applicable to a wide range of flow scenarios.


Authoriser (e.g. PI/supervisor)