Data-based flatness prediction and optimization in tandem cold rolling

In cold rolling process, the flatness actuator efficiency is the basis of the flatness control system. The precision of flatness is determined by the setpoints of flatness actuators. In the presence of modeling uncertainties and unmodeled nonlinearities in rolling process, it is difficult to obtain efficiency factors and setpoints of flatness actuators accurately. Based on the production data, a method to obtain the flatness actuator efficiency by using partial least square (PLS) combined with orthogonal signal correction (OSC) was adopted. Compared with the experiential method and principal component analysis method, the OSC–PLS method shows superior performance in obtaining the flatness actuator efficiency factors at the last stand. Furthermore, kernel partial least square combined with artificial neural network (KPLS–ANN) was proposed to predict the flatness values and optimize the setpoints of flatness actuators. Compared with KPLS or ANN, KPLS–ANN shows the best predictive ability. The root mean square error, mean absolute error and mean absolute percentage error are 0.51 IU, 0.34 IU and 0.09, respectively. After the setpoints of flatness actuators are optimized, KPLS–ANN shows better optimization ability. The result in an average flatness standard deviation is 2.22 IU, while the unoptimized value is 4.10 IU.