In this paper multivariate data analysis performed on a new multi-camera system implemented to zinc flotation circuit of Pyhsalmi mine is presented. Image-analysis-based system was developed to take advantage of the commonly known fact that the changes in the flotation process are reflected to the visual appearance of the froth surface. Most often the changes are seen first in the froth and later in other process measurements.
The image variables are already utilised in closed loop control and considerable savings are achieved with the current setting, but the results coming from the separate cells are used individually. This is clearly not the optimal way to take advantage of the many measurements coming from different locations of the process. A better approach would be for example to use the measurements coming from the early stages of the process to predict the behaviour (or improve the quality of measurements) in the later stages. The methods described in this paper are aimed to improving the situation in this sense.