Soft sensing of particle size in a grinding process: Application of support vector regression, fuzzy inference and adaptive neuro fuzzy inference techniques for online monitoring of cement fineness

Use of soft sensors for online particle size monitoring in a grinding process is a viable alternative since physical sensors for the same are not available for many such processes. Cement fineness is an important quality parameter in the cement grinding process. However, very few studies have been done for soft sensing of cement fineness in the grinding process. Moreover, most of the grinding process modeling approaches have been reported for ball mills and rarely any modeling of vertical roller mill is available. In this research, modeling of vertical roller mill used for clinker grinding has been done using support vector regression (SVR), fuzzy inference and adaptive neuro fuzzy inference(ANFIS) techniques since these techniques have not yet been largely explored for particle size soft sensing. The modeling has been done by collection of the real industrial data from a cement grinding process followed by data cleaning and a structured method of dividing the data into training and validation data sets using the Kennard-Stone subset selection algorithm. Optimum SVR hyper parameters were determined using a combined approach of analytical method and grid search plus cross validation. The models were developed using MATLAB from the training data and were tested with the validation data. Results reveal that the proposed ANFIS model of the clinker grinding process shows much superior performance compared with the other types of model. The ANFIS model was implemented in the SIMULINK environment for real-time monitoring of cement fineness from the knowledge of input variables and the model computation time was determined. It is observed that the model holds good promise to be implemented online for real-time estimation of cement fineness which will certainly help the plant operators in maintaining proper cement quality and in reducing losses. (C) 2014 Elsevier B.V. All rights reserved.