Contribution to the study and design of advanced controllersapplication to smelting furnaces

Resumen

In this doctoral thesis, contributions to the study and design of advanced controllers and their application to metallurgical smelting furnaces are discussed. For this purpose, this kind of plants has been described in detail. The case of study is an Isasmelt plant in south Peru, which yearly processes 1.200.000 tons of copper concentrate. The current control system is implemented on a distributed control system. The main structure includes a cascade strategy to regulate the molten bath temperature. The manipulated variables are the oxygen enriched air and the oil feed rates. The enrichment rate is periodically adjusted by the operator in order to maintain the oxidizing temperature. This control design leads to large temperature deviations in the range between 15ºC and 30ºC from the set point, which causes refractory brick wear and lance damage, and subsequently high production costs. The proposed control structure is addressed to reduce the temperature deviations. The changes emphasize on better regulate the state variables of the thermodynamic equilibrium: the bath temperature within the furnace, the matte grade of molten sulfides (%Cu) and the silica (%SiO2) slag contents. The design is composed of a fuzzy module for adjusting the ratio oxygen/nitrogen and a metallurgical predictor for forecasting the molten composition. The fuzzy controller emulates the best furnace operator by manipulating the oxygen enrichment rate and the oil feed in order to control the bath temperature. The human model is selected taking into account the operator' practical experience in dealing with the furnace temperature (and taking into account good practices from the Australian Institute of Mining and Metallurgy). This structure is complemented by a neural network based predictor, which estimates measured variables of the molten material as copper (%Cu) and silica (%SiO2) contents. In the current method, those variables are calculated after carrying out slag chemistry assays at hourly intervals, therefore long time delays are introduced to the operation. For testing the proposed control structure, the furnace operation has been modeled based on mass and energy balances. This model has been simulated on a Matlab-Simulink platform (previously validated by comparing real and simulated output variables: bath temperature and tip pressure) as a reference to make technical comparisons between the current and the proposed control structure. To systematically evaluate the results of operations, it has been defined some original proposals on behavior indexes that are related to productivity and cost variables. These indexes, complemented with traditional indexes, allow assessing qualitatively the results of the control comparison. Such productivity based indexes complement traditional performance measures and provide fair information about the efficiency of the control system. The main results is that the use of the proposed control structure presents a better performance in regulating the molten bath temperature than using the current system (forecasting of furnace tapping composition is helpful to reach this improvement). The mean square relative error of temperature error is reduced from 0.72% to 0.21% (72%) and the temperature standard deviation from 27.8ºC to 11.1ºC (approx. 60%). The productivity indexes establish a lower consumption of raw materials (13%) and energy (29%).