A Convolutional Neural Network for Classification of Froth Mobility in an Industrial Flotation Cell

Abstract

ABSTRACT Froth flotation is widely used in the resource industry as a particle separation process. The performance of the flotation process is significantly affected by the mobility of the froth phase. Despite its importance, little work has been done to develop a simple and robust method for indicating froth mobility. In the present study, a simple method to monitor froth mobility was developed using a readily available web-camera to take images and a convolutional neural network (CNN) model classifying the images mainly based on the degree of motion blur. The CNN model was trained with a newly built froth image dataset, comprising froth images taken near the overflowing lip of an industrial flotation cell at a wide range of operating conditions using the web-camera. It was found that the trained model could correctly classify 98% of the froth images into one of three categories: low, medium, and high mobility. The froth mobility determined by the trained CNN model was in good agreement with the one analyzed with a commercial software. A potential application of the present method for indicating flotation performance was illustrated.