Advanced deep learning for dynamic emulsion stability measurement. (January 2022)

• Record Type: Journal Article
• Authors: Patil, Amit; Sægrov, Bendik; Panjwani, Balram
• Is Part Of: Computers & chemical engineering. Volume 157(2022)
• Publication Date: 2022-01
• Subjects: Surface chemistry -- Oil-water emulsions -- Interfacial stability -- Pattern recognition -- Faster R-CNN -- Computer vision -- Hough transform -- Circle detection; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2021.107614
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: Experimental study of Exxsol D80 oil-water emulsion stability in a stirred tank cell. Development of Faster R-CNN training data by manual correction of filtered Hough detection results. Performance analysis of Faster R-CNN in relation to filtered Hough method. Dynamic relaxation time evaluation for different Exxsol D80 oil-water systems alongside droplet stability criteria. Abstract: Microscopic probes are used to visualise dispersions in fluid-fluid systems to characterize dispersion behaviors. An advanced multi-stage filtered Hough method and a deep learning based neural networking method (Faster R-CNN) has been compared for droplet detection performance on an image set. The comparison was made with respect to an independent manual analysis. A test analysis for droplet detection performance analysis was prepared. The Faster R-CNN method performed better relative to advanced filtered Hough method. The Faster R-CNN had within 12 % measurement error with respect to the visual or manual detection standard in terms droplet size measurement. This standardized method was utilized to analyze Exxsol D80 oil and water emulsions for evaluating droplet stability parameters. Accurate droplet stability coefficients in the inertial sub-range was evaluated alongside dependency on dispersion phase fraction. The droplet relaxation time scales for the Exxsol D80 oil and water system has been measured and reported where possible.