Recognition of analogous oil droplet attached to transparent pipe wall

Abstract

PTV is an active study method of oil-water two-phase flow characteristic based on photogrammetry. It has advantages of undisturbed, no-contact and high measurement accuracy which directly related to the image quality. However, analogous oil droplet attachments on inner transparent pipe wall are often recorded as part of image, thus reducing measurement accuracy. To overcome the obstacle, it is necessary to identify and locate the outline of analogous oil droplet attachments. Extracting color and motion characters of oil-water two-phase flow images as features for clustering and applying K-means algorithm to identify and locate the outline of the analogous oil droplet attachment. K-means algorithm's clustering result is greatly affected by initial clustering centers and outlier data in practical applications, so Isolation Forest is adopted to improve K-means algorithm. The new algorithm proposed in this paper is called ILF-Kmeans. Simulation and experiment verification are carried out on ILF-Kmeans algorithm. Simulation results show that ILF-Kmeans algorithm has better clustering effect and higher identification accuracy than K-means algorithm; Experiment results show that measurement accuracy of PTV based on ILF-Kmeans to measure the oil phase velocity of oil-water two-phase flow increases by 4.25 %.