Abstract
This paper presents a novel algorithm for a non-intrusive approach to bubble detection without prior training of detection models. In a two-phase flow, one phenomenon of great interest to scientists is bubble detection which provides scientists with parameters of interest such as bubble density, bubble size, bubble volume and bubble velocity which are used to subsequently compute void fraction, quality of the mixture and a host of other parameters. There have been several approaches to detect bubbles including intrusive and non-intrusive methods. One non-intrusive approach is the use of artificial intelligence which involves bubble detection by the use of a trained computer model on a huge dataset of images. However, the challenge to this approach is obtaining enough images to make up the dataset to be used for training and also computational power involved in the training. This work looks at a novel approach to bubble detection which does not involve training a model thereby reducing the computational cost involved. The algorithm involved in this work takes a video feed which could be live or prerecorded as input data, extract the background and use the information from this data to detect the bubbles.
