Soft Sensor for Online Prediction of Cement Fineness in Ball Mill

This paper describes the design and implementation of a soft sensor based on backpropagation neural network model to predict the cement fineness online in a ball mill. The input variables of these models were selected by studying the cement grinding process. The fineness results of laboratory tests were collected to obtain the output variable. This paper introduces a procedure for the extraction, analysis, treatment, and cleaning of raw data received from the factory, which provided a low prediction error. Estimating this variable in real-time can be extremely useful for maintaining the desired fineness during the cement grinding process, which will also allow a significant increase in the system energy efficiency. The model that presented the highest performance and the ability to predict fineness was selected to implement as a soft sensor. The developed system was tested in a cement mill grinding process, and good results were achieved, demonstrating the ability to provide information about the variables previously obtained only through offline laboratory tests.