On neural network based online tuning of rolling mill drive armature current Pl-controller parameters

Abstract

The main tendency of rolling mills drives modernization is outdated analog control systems replacement with new digital ones. At the same time, a power section of the drive remains the same. This gives an opportunity to develop and implement adaptive digital control system for such plants in order to take into consideration their high nonlinearity. Having considered different types of such systems, the ones which does not change the well-accepted PID control algorithm, but tune its parameters KP, KI, KD are believed to have the highest chances to be applied into industry. The main aim of this research is to develop and apply a neural tuner to adjust armature current PI-controller parameters of a two-high reversing rolling mill drive online. The neural tuner consists of 1) a neural network calculating K P and K I values and trained online with the help of backpropagation method and 2) a rule base, which conditions is used to estimate transient quality, whereas consequences are values of a learning rate of the network output layer neurons. Modeling experiments are conducted with a model of the plant under consideration. They are used to check the ability of the tuner to: 1) adjust the PI-controller parameters back to optimal values; 2) cope with an armature winding parameters drift. Such experiments are conducted both for the systems with the neural tuner and a conventional PI-controller. As far as the second type of experiments is concerned, the system with the neural tuner is able to achieve 2% energy consumption decrease comparing to a conventional PI-controller based control system.