Constrained Multivariable Control of a Pilot Flotation Column at the Laronde Concentrator

"A 15 cm diameter by 732 cm height, fully automated pilot column was installed in the Agnico-Eagle’s Laronde concentrator (Québec), for evaluating the performance of newly developed sensors and advanced process control algorithms. For eventual comparison purposes, the pilot column is continuously fed with the same slurry feeding the copper-circuit third cleaner column. It is equipped with electrical conductivity-based sensors for estimating the three variables used in the multi-variable control algorithm under evaluation: froth depth, fraction of wash water underneath the interface and gas hold-up. Froth depth is regulated using a PI controller while a predictive controller is formulated as the minimization of the tracking errors of gas hold-up and bias rate, keeping several operating constraints between their practical upper and lower limits. In particular, an unreachable high set point of gas hold-up can be selected in order to maximize the bubble surface area available for particle collection, while maintaining bias rate above a minimum value required for froth cleaning. The proposed strategy may indirectly help optimizing flotation column operation by using these secondary variables, without the implementation of a real-time optimization layer.INTRODUCTION Flotation columns have already been used for more than 30 years, mainly in cleaning stages, as a result of their froth washing capabilities. Despite their widespread usage, the measurement of some pertaining variables and their use for automatic control is rather scarce. Among the most important variables directly related to the process metallurgy (concentrate grade and recovery), froth depth (hf) is the only one normally monitored and controlled. Even though commercial sensors are available, gas hold-up (eg), a variable directly related to concentrate recovery, is seldom measured and never controlled. Another important variable, bias rate (Jb) responsible for concentrate cleaning, is simply not measured at all, as there is no commercial sensor available. Therefore, besides froth depth, continuous monitoring of process variables is limited to flow rates (gas, feed, tailings, wash water), i.e. to manipulated variables. Some new available sensors have been industrially tested, e.g. the Sonar-TracTM for eg (O'Keefe, Viega & Fernald, 2007) and the McGill hold-up sensor (Tavera & Escudero, 2002). Others, for bias rate and frother concentration, have been successfully validated at laboratory scale by Laval’s LOOP researchers (Maldonado, Desbiens, del Villar & Chirinos, 2008b; Maldonado, Desbiens, del Villar & Aguilera, 2010a). Some devices have been extensively used for monitoring bubble size and space distri-bution, e.g. McGill bubble viewer (Gomez & Finch, 2007) and others."