Comparison of Various Models Implemented Using Linear and Non-Linear SystemIdentification Methods for Cement Grinding Process Using Vertical Roller Mill

Vertical Roller Mill (VRM) has been promptly employed and extensively used for clinker and cement grinding for past 35 years, because of its higher efficiency and low power consumption. VRM is a highly nonlinear, time delay and composite industrial process with several process loops having a robust coupling between them. Time series data is collected from a real time cement plant with grinding capacity of 170-300 tph for system identification purpose. The collected industrial data is normalized and preprocessed using a moving average window filter to remove odd samples; trends and means are removed for outlier elimination and missing values, then the data is divided into an estimation data and a validation data. In this paper linear and nonlinear system identification methods are applied to identify the VRM model. Finally, this paper proposes an accurate VRM model for industrial requirement to adopt modern control techniques like Model Predictive Control (MPC) and Adaptive Control (AC) for cement grinding in order to improve its automatic control level.