Estimación de modelos estáticos para variables de calidad en el circuito de molienda de UCEM, planta industrial Guapán

Abstract

Cement grinding is the responsible for the fineness characteristics of the final product, typically measured by two quality variables: specific surface or Blaine (cm2/g), and the percentage passing the sieve No. 325. It is a common practice in cement plants to perform laboratory tests of the fineness of the cement every two hours, leading to a gap of what produces an interval of two hours of blind operation regarding cement quality. As a response to this problematic, in this paper, it will be developed models of prediction of the quality variables, which allows minimization of the final product variable or increased the energy efficiency. The implemented models are: statistics, autoregressive, neuronal, and neuro-fuzzy. A statistical analysis of the database is performed, provided by the UCEM-Guapán factory, corresponding to the operation and quality variables of cement grinding. The selected variables for the development of the models are made through the treatment and analysis of the correlating data. The validation takes place in with operation values in the plants of UCEM. For the selection of the best model, it uses the residual error technique. The model with the best results is implemented in the LabVIEW software, for both the variable of Blaine and for percentage passing the sieve.