(383f) Real Time State Estimation of Froth Flotation Processes Using the Extended Kalman Filter and Image Processing

Froth flotation is used to extract valuable minerals using differences in the hydrophobicity of the valuable mineral and other components of the ore. It is one of the most widely used processes for separation in the mineral industries. Flotation is a combination of various complex processes occurring in the pulp and froth phases, and a large number of factors affect the recovery of the valuable minerals. This makes monitoring and control of the flotation process a difficult task. The main goal in froth flotation is to maximize grade and recovery of the desired mineral, while maintaining stable and upset-free operation (Fuerstenau, 2007). However, with highly varying feed properties and other related operating parameters, potentially leading to poor froth stability, which can result in reduced grade and/or recovery of the desired mineral. To remedy this, industrial practice is to apply closed-loop feedback control to the process. The control strategy typically tries to control the froth structure by controlling bias, froth depth and gas hold-up by manipulating variables such as the air and wash water flow rates, and reagent addition rate. Dynamic models built for this purpose are largely empirical (e.g. Bouchard et al., 2005, 2009; Desbiens et al., 1994). Real-time measurement of grade and the corresponding calculation of recovery is often achieved by using an X-ray fluorescence (XRF) analyzer (e.g. Remes et al., 2007) or by image analysis of the froth (e.g. Liu and MacGregor, 2008; Bartolacci et al., 2006; Aldrich et al., 2010).

However, in the absence of a fundamental model, there is no way of determining which of multiple possible disturbances has affected the process; also, the empirical models are typically linear, and are only valid in small zones of operation around a steady-state operating point; thus, larger deviations may cause the models to be inaccurate. Also, identification of such empirical models is relatively difficult for disturbance-driven processes, and these models do not provide any physical insight into the process and its condition. If a fundamental model can be developed for froth flotation and reconciled with data in real time, a better understanding of the process along with more effective recovery control can be achieved.  In this work, we use the Extended Kalman Filter (EKF) for state and parameter estimation. Since the development of fundamental models for froth flotation is a challenging task, we employ the approach of developing a hierarchy of dynamic models at varying levels of complexity. These include models which approximate the process of recovery using first order mass action kinetics, and more detailed models that model attachment and detachment processes in the pulp phase separately, along with transfer of material between the pulp and froth phases. In addition, we include models based on distributions of particles in terms of their size and hydrophobicity, and account for entrainment. Within this hierarchy of models, we provide information on parameter estimability and state observability with each of these models, thus generating insight into the question of which model is most appropriate for on-line updating, given that different models have different numbers of parameters to estimate.

Along with a model, the EKF requires real-time measurements in order to update the estimates of the model states and parameters. As most froth flotation variables are not directly measurable, especially in real time, image processing techniques are applied to generate on-line measurements and continuously monitor froth properties. This is accomplished by using “Visio-Froth”, an imaging package by Metso, which uses a camera, laser and LED light to obtain measurements of froth velocity, bubble size distribution, color and texture along with other bubble properties.  Watershed techniques are used to delineate bubble contours. Grade and recovery are calculated based on calibration of the bubble properties against off-line measurements.

The estimation and soft sensing algorithm is tested on a continuous flotation cell constructed at the University of Alberta for the separation of Pb-Zn sulfide ore using Xanthate and MIBC as collector and frother, respectively. Guidance is provided on suitable complexity of models for on-line estimation, and suggestions for off-line measurements to identify other parameters of the models are also provided.

References:

M.C. Fuerstenau, G. Jameson and R.-H. Yoon, Froth flotation: A century of innovation, Society for Mining Metallurgy & Exploration (2007).

J. Bouchard, A. Desbiens and R. del Villar, Recent advances in bias and froth depth control in flotation columns, Minerals Eng., 18, 709-720 (2005).

J. Bouchard, A. Desbiens, R. del Villar and E. Nunez, Column flotation simulation and control: An overview, Minerals Eng., 22, 519-529 (2009).

A. Desbiens, D. Hodouin, K. Najim and F. Flament, Long-range predictive control of a rougher flotation unit, Minerals Eng., 7, 21-37 (1994).

A. Remes, K. Saloheimo and S.-L. Jamsa-Jounela, Effect of speed and accuracy of on-line elemental analysis on flotation control performance, Minerals Eng., 20, 1055-1066 (2007).

C. Aldrich, C. Marais, B.J. Shean and J.J. Cilliers, Online monitoring and control of froth flotation systems with machine vision: A review, Int. J. Mineral Processing, 96, 1-13 (2008).

G. Bartolacci, P. Pelletier Jr., J. Tessier Jr., C. Duchesne, P.-A. Bosse and J. Fournier, Application of numerical image analysis to process diagnosis and physical parameter measurement in mineral processes – Part I: Flotation control based on froth textural characteristics, Minerals Eng., 19, 734-747 (2006).

J.J. Liu and J.F. MacGregor, Froth-based modeling and control of flotation processes, Minerals Eng., 21, 642-651 (2008).