Non-linear modeling of cement grinding process

The cement industry is an energy-intensive industry like chemical, iron-steel, petrochemicals and paper industries and has high-energy demands. Today, the cement production capacities of many cement producers are limited to their energy consumption. It is possible to increase the production capacity but only if the energy consumption of process is reduced. When all of the cement production processes of the cement industry are examined, it can be seen that the most energy consumption is in the cement grinding process but it is a stochastic process due to the presence of several sources of noise and fluctuations. The quality of cement varies depending on many operating and process parameters, which may vary from system and time. These parameters include the ball size, feed size distribution, feed rate and tonnage, material hardness and grain size distribution and the speed of air in the mill and separator. The desired value of cement quality and energy consumption is possible by taking the most suitable values of these variables and by controlling the best way of the system. It is not possible to derive a deterministic model of the system because of the parameters that are covered by the bet, so experimental methods such as system identification stand as the only alternative of the best model of the system. In this study, the modeling of the cement grinding process by linear and non-linear methods is investigated in the cement industry. The cement grinding process was first described by the ARX and ARMAX models, which are linear parametric model structures that must be applied first in system diagnosis. The system was modeled with NARX, Hammerstein-Wiener and artificial neural networks (ANN), respectively, with the result being considered inadequate. In both linear and nonlinear modeling studies, the system was modeled as a whole. Since there are five inputs and one output in the system model, unknown parameters between ten and sixteen must be estimated for each model of the cement grinding process. Because of the parameters and the size of the solution space, Genetic Algorithm (GA) Optimization Method has been used for optimization of these parameters. In the first part, general information is given about the thesis, the aim of the thesis and previous literature studies are given. The second part gives general information about cement, cement production stages and energy consumption distribution in production stages. At the end of the chapter, the cement grinding process, in which modeling studies are carried out during the thesis work, is introduced. In the third and fourth chapters, the description of linear and nonlinear dynamical systems is mentioned. As the linear system diagnosis models, AR, ARX, ARMAX, BJ and OE models are explained. In the fourth part, nonlinear system identification models are given information about NARX, Hammerstein-Wiener and artificial neural networks (ANN). In the fifth section, we give information about Optimization Method of Genetic Algorithm (GA), which is used in optimization studies of parameters. In the sixth chapter, before the cement grinding process modeling studies are started, the collection, preparation and performance criteria of the models are explained. In the seventh chapter, the cement grinding process is investigated according to different modeling patterns. Results related to models are obtained and compared. In the last chapter, chapter 8, briefly the results obtained during the study are brought together and summarized. In addition, these results are compared and the contribution of working in a summary is tried to be explained.