MEASUREMENT OF GEOMETRICAL PARAMETERS OF THE CRUDE-OIL/WATER INTERFACE PROPAGATING IN MICROFLUIDIC CHANNELS USING DEEP LEARNING TOOLS

Abstract

This paper reports the results of the application of some software tools based on deep learning models, on the processing of microscopic images of the interface between crude oil and water, while propagating in microfluidic channels. The U-Net deep learning model is used to classify the pixels of the crude oil and separate them from the rest of the pixels (semantic segmentation). This has allowed for the automatic measurement of some geometric parameters of the meniscus, making possible the processing of large amounts of images. Live videos of the meniscus have been recorded while the water propagates in the microfluidic guides previously filled with crude oil, and then the frames (images) from the video have been extracted and processed. In this way, we were able to consider the information about time and also study the dynamic behavior of the geometric parameters. Among the geometric parameters that it is possible to measure, the angle between the meniscus and the walls of the propagation channel were chosen. The angle measured from the propagation images was compared to the contact angle measured in a static regime, with the method of the sessile drop.