Modeling of Industrial Limestone Grinding Process Using Regression Models and Neural Networks

Abstract

The article discusses the problem of mathematical identification of the limestone grinding process in an industrial roller-bowl mill, with particular emphasis on assessing its energy consumption. This is a key stage in the optimization of the grinding process because the development of effective indicators models to assess the technological process guarantees its effective optimization. Therefore, technological and energy indicators were proposed that assess the limestone grinding and classification process, based on which regression models and neural networks were built. These models show the relationships between the process evaluation indicators and the energy and technological data obtained during a factorial experiment at the milling facility. The industrial experiment aimed to examine the nature and dynamics of changes in process parameters. During the experiments, the controllable technological parameters of the tested process were changed within a wide and technologically acceptable range. Process identification was based on data from control and measurement devices and the process control system, which enabled full mathematical modeling of the process and explanation of the interdependence of process parameters. Regression and neural network models were determined, allowing an effective identification of the operation of the entire limestone grinding and classification process. The models were subjected to statistical diagnostics which aimed to verify their construction's correctness and compare the operation's effectiveness. Regression models were more practical in assessing the grinding process, while neural networks were more useful in control, thanks to the possibility of automatic learning and updating in dynamically changing process conditions.