Recursive multimodel partial least squares estimation of mineral flotation slurry contents using optical reflectance spectra

In mineral flotation the X-ray fluorescence (XRF) grade measurements of the slurries typically give the most important on-line information on the state of the flotation process. It has been shown that the visual and near-infrared (VNIR) reflectance spectrum measurements of certain mineral slurries can be used to complete the sparse XRF slurry content information. This study focuses on the chemometrical analysis of the VNIR spectrum of the slurries and presents a new partial least squares (PLS)-based recursive multimodel approach with local orthogonal signal correction (OSC) for predicting the slurry contents. The advantage of the presented approach is that it can recursively adapt to real process data variations in normal operating conditions, and is still able to remember the rare process failure situations with notably different content values.