Acquisition and analysis of floc images by machine learning technique to improve the turbidity removal process

Abstract

This article reports on the implementation and use of a floc image acquisition and analysis system in a pilot water treatment plant to remove kaolin turbidity with a coagulant and flocculant. The system is based on the Hausdorff dimension (df) of the images and is used to obtain information about the image texture and to ensure that the flocs could be removed by the filtration system, and to use df values for corrections of the dosage of both chemical agents via signals with pulse width modulation that feed and control dosage pumps during treatment, ensuring a continuous adjustment for changing water conditions, which allows for a close on-site process control and a rapid response to changes in the quality of the effluent.