Online Prediction Of The Actual Melt Chemistry In An Ausmelt Converter Using A Thermodynamic - System Identification Hybrid Modeling Technique

Lance based copper-nickel matte converters have become an established technology in the field of base metal sulphide smelting and converting. An Ausmmelt® converter is currently being constructed at Anglo Platinum in South Africa to convert a nickel-copper rich matte containing platinum group metals (PGM's). The accurate prediction of the matte chemistry is required to control the composition at the endpoint which influences segregation of PGM's during the subsequent slow cooling and magnetic separation processes. A hybrid modelling approach is introduced in this paper that incorporates both the effect of thermochemistry and system dynamics. It will be shown how the model relates to fundamental conservation laws, while still providing enough flexibility to adapt to changing operating conditions. It will also show how off-line modelling can be incorporated into a robust on-line model. Based on the data from many pilot plant trials, the model prediction performance is shown to be within the accuracy requirement set by industry and therefore constitutes a reliable basis for model based control. The final generic model structure takes on the form of an thermodynamically integrated ARMAX hybrid model. The final model has the advantages of being dynamic, adaptive, robust and it can be related to thermochemical fundamentals.