Real-time hybrid predictive modeling of the Teniente Converter

Abstract

The Teniente Converter (TC) is an important technology for smelting and converting copper concentrates. Due to the complexities of its autogenous operation, which combines continuous input flows with intermittent product extraction, efforts to model and control this dynamic have only been partially successful. This work presents a model of TC operation that combines phenomenological equations with empirical expressions and includes both continuous and discrete variables. Six different operating modes are described. To improve predictive capabilities, an adaptive compensation technique takes account of changes in concentrate mix characteristics. The model predicts five important variables: white metal copper concentration, slag magnetite concentration, white metal temperature, and white metal and slag levels. It is calibrated and validated with real operating data from Codelco’s Potrerillos smelter. Once the model is implemented, it delivers online estimates of process variables that are measured intermittently. Based on the model, a hybrid predictive controller is developed to control the quantities and characteristics of the TC products as a function of the number of dependent units such as Peirce-Smith converters and slag-cleaning furnaces. The controller’s performance is evaluated via simulation under a variety of operating conditions. It is currently employed in an open-loop configuration to provide suggestions on set points for the local flow controllers and on the product extraction sequences.