Oil sand screen modelling using partial least squares regression

In the oil sands industry, screening is a critical part of the mining process. Oil sand companies use screens to separate oversized lumps from the surrounding oil-rich sand before it enters the oil extraction process. However, during unknown screening conditions, varying proportions of the sand will pass over the screens, resulting in unexplained variations in screening performance. Typical screen performance evaluation involves extensive sampling of the feed stream in order to obtain comprehensive data of typical screening conditions. In this work, we present a general methodology for using non-traditional plant variables, such as geological information, to develop an oil sand screen model when traditional screening variables, such as the feed particle size distribution, are difficult to obtain. A screen model that uses feed rate, water and geological variables to explain screening performance variation is developed using partial least squares regression. The model’s behaviour is compared with a model that uses only the feed rate. Results show an average 25 percent reduction in RMS error and an average 0.34 increase in the adj-R
²
over the feed-rate-only model. This is the first known study to consider nonconventional water and geological variables, other than feed rate, for modelling an online oil sand screen.