Pergamon 0892,-6875(98)00059-4 Minerals Engineering, Vol. 11, No. 8, pp. 739-748, 1998 © 1998 Elsevier Science Ltd All fights reserved 0892-6875/98/$ - see front matter FUZZY SUPERVISORY CONTROL OF FLOTATION COLUMNS L.G. BERGH, J.B. YIANATOS and C.A. LEIVA Chemical Engineering Department, Santa Maria University, Valparaiso, Chile E-mail: lbergh @itata.disca.utfsm.cl (Received 4 February 1998; accepted 18 May 1998) ABSTRACT The application of fuzzy logic to supervise a distributed basic control in a flotation column is discussed. The control strategies were studied and tested in a dynamic simulator of the process. Two control strategies were developed and tested to manage three basic distributed controllers: an expert supervisor, mainly based on rules following a binary logic, and a fuzzy supervisor. The objective function to be optimized was to keep the concentrate grade in a high band, subject to maintaining the process recovery over a minimum value. The supervisor takes into account the present state of the gas flowrate, the froth depth and the wash water flowrate, to make a decision. Simulated results are discussed. © 1998 Elsevier Science Ltd. All rights reserved. Keywords Column Flotation, Artificial Intelligence, Expert Systems, Process Control INTRODUCTION Flotation columns are now used worldwide as efficient cleaning stages in a large number of sulfide mineral concentrators. More degrees of freedom in their operating variables have led to large variations in metallurgical performance and therefore to much scope for improving their control. In this process, stable operation and consequent consistent metallurgical benefits can only be obtained if basic distributed control systems are implemented. In Figure 1 a representation of a typical flotation column instrumentation and control system is shown. In general, at least wash water and air flowrates and froth depth are measured on line, and tailings, air and wash water flowrates are manipulated. On line analysers, tailings and feed flowrates and some other measurements are often incorporated into the system when a supervisory control strategy is implemented on top of a distributed control system. The primary objectives, as indices of process productivity and product quality, are the recovery and the concentrate grade. However, their on-line estimation usually requires a significant amount of work in maintenance and calibration of on-stream analysers, in order to maintain good accuracy and high availability. Therefore, it is common practice to control secondary objectives, such as the froth depth, the 739