On neural tuner application to adjust speed P-controller of rolling mill main DC drive

Abstract

Mechanics nonlinearity of such a complicated plant as rolling mill leads to deterioration of transients quality for a main drive speed control loop. A main reason for that is linearity of P-control algorithm used for this loop. That problem can be overcome by controller parameter adaptation algorithms usage. In this research we propose to adjust speed P-controller online using a neural tuner. It is implemented as a combination of an artificial neural network, which output is KP, and a rule base. The base reflects an automation engineer experience and allows to determine situations when the controller is to be adjusted, i.e. when to train the network. It also contains appropriate learning rate for the neurons of the network, which takes into account the nonlinearity of the plant. Numerical experiments are conducted using the model of the considered plant main drive. They are divided in two parts. The first one includes experiments, when the tuner task is to tune nonoptimal speed controller parameter back to the calculated one. The second one is made for the case of plant mechanical part parameters drift. The results show that neural tuner application allows to decrease energy consumption of considered DC drive by 1.9% in comparison with P-controller without adaptation.