for
Semi-autogenous Grinding Mills"
T.A.APELT
2 May 2007
A thesis submitted for the degree of
Doctor of Philosophy
of the University of Sydney
Univeristy of Sydney
NSW 2006 Australia
Inferential models are presented for SAG mill total and ball charge fractions, fresh feed rate and size distribution and discharge rate and size distribution, according to the research objectives. Inferential models are also introduced for the SAG mill recirculating load, rock charge, and total feed rate and size distribution and the primary cyclone underflow split fraction. The inferential models could be utilised as a measure of SAG mill performance or indication of the primary grinding circuit operating conditions.
Important model parameters are highlighted by sensitivity and uncertainty analyses presented for the inferential models of the SAG mill charge levels, discharge rate and size distribution and the fresh feed rate and size distribution. The results indicate that the mill charge level estimates obtained from the mill weight measurement contain the least uncertainty and are therefore the recommended choice for charge level estimation. Uncertainty may be minimised through the utilisation of the best available mill weight measurement and periodic measurement of mill inside diameter and length. The non-linear nature of the powerdraw model results in a high degree of uncertainty in the total charge estimates obtained from the powerdraw measurement.
Uncertainty in the mill discharge estimates may be minimised by ensuring the accuracy of the oversize crusher and primary cyclone feed instrumentation is maintained through regular calibration and the periodic measurement of the SAG mill discharge screen aperture size and the process water specific gravity.
Uncertainty in the SAG mill fresh feed estimates may be minimised by ensuring the accuracy of the oversize crusher and primary cyclone feed instrumentation and the crusher gap setting. The SAG mill discharge grate parameters: pebble port size and relative open area fraction of the pebble ports, also strongly influence the size estimates and hence, should be fitted with due care.
Combined state and parameter estimation formulations are developed for the SAG mill, according to the _nal research objective. The formulations incorporate novel models of the SAG mill ball charge and protective shell lining and mill weight and discharge measurements. The system is detectable although not completely observable. Filter tuning parameter selection is crucial to formulation performance. The results indicate that the superior formulation incorporates a size-by-size SAG mill discharge measurement model which provides better capacity to estimate important mill discharge grate parameters. The formulations also provide a suitable, positively supportive context for the inferential models presented in this research.
Plant data was sourced and the inferential measurement models are validated against it. The results confirm that the inferential measurement models are valid. Results analysis also reveal the potential for utilisation of the inferential measurement models in a process monitoring/diagnostic capacity.
The models are analysed further in terms of sensitivity and their inherent nature, which illustrates a limitation in the feedsize estimate model that should be noted when dealing with model results. Mill weight and powerdraw contours and a SAG mill operating curve are generated and presented in charge fraction and charge fraction - kilowatt•tonne spaces. The contours and operating curve are discussed in the context of mill charge control.
A multi-variable, model predictive controller is developed that utilises the inferential measurement models for setpoint and constraint-control. The development of the controller incorporates transfer function relationships and the description of the SAG mill variables of powerdraw, weight, rock charge and ball charge as integrating variables. The performance of the controller is documented and assessed.
Further research recommendations centre on further model validation against industrial data, further parameter sensitivity analyses, simulation model development and the investigation of the potential use of the inferential measurement models in a process monitoring capacity. Investigation of the dependence of the operating curve on ore hardness and the investigation of control actions of a real controller in relation to the operating curve are other key areas to progress the research. Implementation recommendations are particularly relevant to Northparkes Mines but also have relevance for the wider minerals processing community.
Increasing the number of inferential measurement models available to industry and the body of knowledge supporting them adds further impetus to the transition taking the control methods of autogenous grinding circuit supervisory control and the on-line use of phenomenological models from the “active” or “emerging” phase to the “mature” phase.
