Application of Multivariate Image Analysis (MIA) to Predict Concentrate Grade in Froth Flotation Processes

"The development of on-stream analyzers (OSA), such as those based on X-ray fluorescence (XRF), has been key to improve metallurgical performance of flotation processes. However, since these analyzers are shared between several streams, they typically provide grade measurements every about 10-15 minutes. In general, this frequency is too long compared to process dynamics and so, implementing an efficient feedback control strategy based on OSA can be difficult. In addition, due to the limited capacity of OSA systems, several important intermediate streams are not analysed, but could provide relevant information about the state of a flotation circuit.To overcome this problem, vision systems have been developed in the past to predict concentrate grade based on froth color – this could be very useful to control flotation performance between OSA measurements. However, applications reported in the literature typically use univariate summary statistics of the full red, blue, and green color distribution of froth images, while RGB images are truly multivariate. Although, in some cases, grade prediction was achieved with reasonable accuracy, one may, by using only summary statistics, miss important information about the state of the flotation process.This paper investigates the application of Multivariate Image Analysis (MIA) concepts to flotation froth images. These techniques are based on multivariate statistical methods such as PCA and PLS, which have the ability to analyze large databases (such as digital images) and summarize the relevant information content of the data into a small number of latent variables, capturing the most important features of the images for concentrate grade prediction. The potential of application of these methods in the mineral processing industry is illustrated using a few industrial case studies."