THE USE OF AN IMAGE PROCESSING BASED SENSOR FOR ON-LINE ANALYSIS OF FLOTATION PERFORMANCE

摘要

Plant operators can qualitatively assess the performance of a flotation process by visual inspection of the surface of the froth formed in flotation machines. This paper describes a computer vision system designed to provide consistent and unbiased descriptors of surface froth characteristics. Using video cameras and personal computers, images of surface froths from industrial flotation plants are acquired and digitized. Efficient image processing techniques based on the Fast Watershed Transform are used to segment the froth images in individual bubbles. Physical descriptors (such as bubble size and shape) are extracted from individual segmented images, while dynamic characteristics (froth velocity) are obtained from sequences of images. The descriptors obtained are used to elucidate the relationship between surface froth properties and plant performance. Once established, these relations can be used to detect deviations from optimal plant performance. Once established, these relations can be used to detect deviations from optimal plant performance and, when used in conjunction with expert control systems, return the plant to the best operating conditions by making appropriate process input changes.