Real-Time Estimation Of Unmeasured Variables In A Semiautogenous Grinding Circuit ? Introduction

It is widely recognized that Semiautogenous Grinding (SAG) circuits require the use of effective process control techniques to achieve their full processing potential and to protect them from the harmful effects of overloads and grind-outs. This requirement is even more essential as the variability of the ore being processed increases. The new Chino Mines Company concentrator located near Santa Rita, New Mexico started operation in June of 1982. At least eight major rock types are mined and milled. Ore hardness, feed size, and specific gravity are quite variable. Typical feed rates per grinding line are 850-900 tph but extremes of 260-1200 tph have been encountered. Normal feed rate variations of between 400 and 1100 tph are encountered on a weekly and sometimes daily basis. Feed size varies with rock type from a maximum of eight inches for the harder ores to two inches for the softer ores. One somewhat unique characteristic of the Santa Rita ore body among major copper mines is the wide variation in ore specific gravity. Specific gravities of the major rock types range between 2.7 and 3.6. The specific gravity variations are due to large changes in magnetite concentration throughout the pit. The result of this is large variations in SAG mill throughput which can have a harmful effect on downstream ball milling and flotation performance. Due to these wide fluctuations in ore characteristics, it was recognized soon after startup that successful operation of the grinding circuit and the plant at large was dependent upon the development of an effective automatic grinding control strategy. Research at the University of Utah has suggested that control based upon critical but difficult or costly to measure variables would improve automatic control performance. A project team made up of the Chino process control staff, Kennecott's Process Control department and the University of Utah Metallurgy Department was established to assess these concepts at Chino. This paper describes the development and on-line plant implementation of an Extended Kalman Filter that uses simple models of the SAG milling process, to predict the volumetric loading, the rock mass, the water mass, and the ball mass within the SAG mills at Chino. The on-line estimation of these parameters is the first step in the development of an integrated SAG mill/ball mill control strategy which is hoped to L eventually maintain the grinding circuit at its maximum capacity, while balancing the grinding load between the SAG and ball mills. Examples of long term estimation with the models and filter are shown for typical operating conditions as well as a SAG mill overload and grind-out.