Análise da influência dos erros de modelagem no desempenho de sistemas de controle

Abstract

The use of model-based control and optimization systems presents high potential to increase the efficiency and profitability of industrial processes. Therefore, the use and interest in these techniques have grown up steadily. Particularly, the performance of model based systems, as predictive controllers, depends on several factors that affect the quality of the estimated model and the selected tuning procedures. It is important to notice that, when dealing with complex industrial processes, the model building process often involves a large number of simplifications. The analysis of the open literature shows that many theoretical and practical issues related to the use of predictive controllers still need to be studied. In particular, the relationship among modelling, tuning and control performance is not completely explained and explored. Thus, in this thesis the relationship between modelling and tuning issues is analyzed, focusing at the controller performance when step response models are used. Based on the theoretical studies, new methods have been proposed for controller design and tuning and for the determination of the model relevance. The proposed methods were then applied in validation examples which showed that it is possible to relate the tuning parameters with quantitative measures of model performances, such as the parameter variances estimated during the model identification phase.