Multivariable Control of a Rougher Flotation Cell

Abstract

This project focussed on the investigation, development and evaluation of a closed loop

control system on a rougher flotation cell that could improve PGM flotation performance.

A PGM rougher flotation cell equipped with Online Stream Analysis (OSA) and machine

vision system (SmartFroth) was used during the investigation. Online measurements

included bubble velocity, average bubble area, bubble colour; Pt, Ni and Cu concentrate

grade and concentrate flow rate and density. Air addition and pulp level was used as the

manipulated variables.

A system identification exercise was performed to model the dynamic responses of all

the available outputs to step changes in the air addition and pulp level. Brief analysis of

the conSistency of the output gains for different initial conditions was done to determine

whether the process for the measured outputs were linear. None of the outputs that were

selected to be controlled exhibited non-linear behaviour so that linear control theory

could be applied for the development of a closed loop controller.

Linear dynamic models of the machine vision outputs were analysed and it was

determined that the biggest operating region for MI MO control exists when the air

addition is manipulated to control the bubble velocity and the pulp level is manipulated to

control the blue bubble colour. The multivariable control application was relatively simple

as the interaction between the two control loops was intermittent. Different decoupling

structures with PI controllers were tested in a simulation environment and It was found

that the difference between the decoupling structures were small with no single

decoupling structure outperforming any of the other structures. LOG and IMC MIMO

controllers were also designed and compared in the simulation environment. All the

controllers performed similarly with the exception of LOG control, which was worse.

Implementation of the designed controllers on site exhibited mixed results. The blue

colour measurement proved to be unreliable while the bubble velocity showed good

control potential as an industrial measurement by manipulating the air addition.

Research should focus on relating machine vision outputs directly to the metallurgical

performance of flotation so that optimal machine vision setpoints can be selected.