Computer Vision in Fluid Mechanics

Master of Science, Graduate Program in Mechanical Engineering
University of California, Riverside, June 2015
Dr. Marko Princevac, Chairperson

Flow visualization is one of the main methods of understanding flow behavior in experimental fluid  mechanics.  Most  of  the  visualization  methods  provide  good  qualitative  information about the flow behavior. With the developments in computer sciences especially in the fields of artificial intelligence and computer vision, image processing has been used as an important tool  in  experimental  fluid  mechanics  to  quantify  the  qualitative  data  obtained  by  flow visualization. This thesis first summarizes various flow visualization methods. One of the most common use of computer vision is in Particle Image Velocimetry (PIV). PIV uses block matching  algorithm  to  detect  motion  of the  fluid  between  two  frames  by  calculating  the displacement of small particle groups. PIV has many downsides and is not applicable in numerous fluid mechanic studies such as two phase flow. Because of the downsides of PIV it is  often  better  to  use  other  visualization  technics  and  more  advanced  image  processing algorithms. Therefore, instead of block matching algorithms we investigate more advanced algorithms such as optical flow measurements that can provide flow properties for variety of applications.  Next,  we  modify  the  existing  algorithms  and  apply  them  to  study  bubble behavior in a turbulent flow. Using the optical flow and edge detection algorithms, perimeter velocity  of  the  bubble,  and  geometric  properties  of  the  bubble  were  obtained  in  various frames.  It  was  found  that  the  turbulent  intensity  of  the  bubble’s  perimeter  velocity  is converging to a common value when the bubble breaks and reaches its smallest possible size. Based on this finding we use bubble movement to evaluate turbulent flow properties in two phase  flow.  Future  work  may  include  integration  of  all  involved  techniques  for  further optimizations of parameters involved.