Nonlinear Model Predictive Control of Industrial Grinding Circuits using Machine Learning

Control problems of engineering interest such as industrial grinding circuit (IGC) are essential for minimizing the energy consumption, maximizing throughput or maintaining product quality to make these processes energy sustainable in future. Detailed physics based models, although provide more insight into the process and yield accurate results, often cannot be used for control purposes owing to the high computational time involved in solving the complicated mass, momentum, energy balance equations used to describe the process. In the current study, the aim is to use optimally designed Recurrent Neural Networks, a type of data-based modeling technique popularly used in the machine learning domain, for modeling transients involved in the IGC and test its effectiveness in set point (SP) tracking under the nonlinear model predictive control (NMPC) framework. SP tracking of throughput (related to the amount of raw materials processed to give products) and recirculation load (related to the energy consumption) is performed. We observe that the developed machine learning based models could effectively perform the SP tracking for nonlinear industrial process like grinding.