The Use of Data Reconciliation as a Soft Sensor in Various Hybrid Modelling and Control Architectures for Pyrometallurgical Furnaces

In this paper various hybrid modelling architectures are proposed for the modelling and metallurgical control of pyrometallurgical reactors. Discussed are architectures in which data reconciliation is combined by thermochemical, neural net and dynamic modelling to improve the models. At the same time the data reconciliation models can act as soft-sensors for estimating unmeasurable data. Cases that are discussed are (i) the modelling of a submerged arc furnace for the production of thermal phosphorous, (ii) the processing of Sn-Pb-Zn flue dust (originating from secondary copper) in a hearth furnace, (iii) modelling of a steel converter, and (iv) dynamic modelling of a flash smelter.