Guest Lecture by Bernard Coetzee and Lidia Auret from Stone three.

From computer vision to concentrator value: Robust and actionable froth and pulp analytics

Bernard Coetzee, Lidia Auret

Flotation froth image analysis has been a topic of academic research and industrial implementation since the early 1990s, while online pulp image analysis is a more recent phenomenon. Algorithmic advances (such as deep learning) have led to improved accuracy in bubble segmentation, while also allowing additional inferences to be made on the operational state of the observed flotation cell. In this talk, we will discuss various practical considerations which ensure that commercial froth and pulp image analysis yields robust and valuable measurements. These considerations include sensor sampling frequency, sensor integrity analytics, operational state estimation, and the implications of these considerations in automated control.