Data-Driven Optimization Control for Safety Operation of Hematite Grinding Process

Grinding particle size (GPS) of a hematite grinding process (HGP) is characteristic of sensitivity to disturbances of ore hardness and size distribution, difficulty in establishing a mathematic model, and absence of measurement, which make the existing operational optimization and control approaches difficult to be applied. This leads to that it is always difficult to obtain a proper ore feed rate and water flow rate during process operation, and consequently, overload fault often occurs. To tackle this problem, based on characteristic analysis of GPS, a GPS prediction algorithm is first developed using an improved neural network (NN), and a data-driven optimization control approach for safety operation of HGP is proposed. The proposed method adopts a two-layer structure, the higher-level operational feedback control, and lower-level basic loop control. The operational feedback control consists of a cascade NN-based loop setpoint optimizer and an overload diagnosis and self-healing controller. The main advantage of the proposed method is that it requires only the available operating data, without knowing the process dynamics. Experiments have been carried out in a self-developed hardware-in-the-loop experiment system with actual data, demonstrating that the proposed method cannot only keep the HGP operation from mill overload fault but also achieve the optimization control with respect to the quadratic performance index of GPS deviation.