Validação da simulação dinâmica das etapas de evaporação e cristalização da produção de açúcar com dados obtidos em plantas industriais.
Jesus, Charles Dayan Farias de
MetadataMostrar registro completo
The central area of the State of São Paulo, Brazil, where the city of São Carlos is located, is one of the largest world producers of alcohol and granulated sugar from sugar cane. As the market become more competitive, the factories seek for new technologies and more rigorous mathematical approaches. This work is inserted in this context, where partnership between university and industry is encouraged. The key steps in the sugar production process are the evaporation of the juice and the crystallization of sugar. The evaporation stage has great influence in the energy balance of the factories but, in spite of that, its automatic and optimized operation is not completely implemented in a large part of the producing units. By its turn, the crystallization step should be monitored very closely because is in this step that the final product is obtained, so quality issues are critical. In this work, the main aim was to develop non-linear dynamic models of these two important operations. The models used industrial data collected from Usina Santa Adélia (evaporation stage) and Usina São Martinho (crystallization stage). The processes of the two factories were accompanied, analyzed and sampled. A steady-state and a dynamic models of the evaporation were developed using the classical approach material and energy balances. Both models were designed in such way that they depend only on few assumptions and on measurements available on-line. The results obtained with the steady-state model fully characterized all streams of the multiple-effect evaporator and allowed to estimate the steam flow rate bled from the equipment. With the dynamic model it was possible to calculate the concentration of the syrup in the last effect. Comparison between the predicted values and the actual industrial data showed that the model generated estimates in the same range of values and reproduced the behavior of the variable. The analysis of the results suggested that more measurements would be necessary to develop a reliable model (in the process control sense) and demonstrated that it is difficult to use the industrial data in the format they are found in the historical data files. To overcome these inconveniences an artificial neural network was developed as a software sensor for the Brix in the last effect. Several topologies were tested and the results of the best ones were very good. Unfortunately, as the performance of the predictions depends on the quality of the training set, it was necessary a real effort to preprocess all available data. The proposed model for the crystallizer (batch vacuum pan) select in the crystallization stage was constituted by a system of nonlinear algebraic and differential equations. Besides the energy and material balances, the model included the population balance, which allowed the calculation of the average crystal size and its coefficient of variation during the strike. Even with severe assumptions and few input data, the model was able to make good predictions of the rajectories of the state variables of the equipment. The analysis of the results also showed that some aspects of the automatic operation affect the stability of the process. Finally, it could be said that the first principles and the neural network approaches could become powerful, useful and reliable tools to model the stages of sugar production if more sensors were installed in the process.