Corresponding author: Eremenko Yurij I., Professor, Stary Oskol Technological Institute named after Ugarov, MISIS National University of Science and Technology (branch), Stary Oskol, 309516, Russian Federation, e-mail: erem49@mail.ru

Received on March 23, 2017
Accepted on April 14, 2017
Rolling production is one of the most energy-consuming sectors of the metallurgical industry. The most powerful plants are the rolling mills producing rough rolling of the steel casts. They are based on the technology of the reverse rolling. It requires variation of the roll mill parameters. Such a variation may also be due to replacement of the worn out rolls. So, the controllers with constant parameters usage results in the deterioration of the transient quality for the rolling mill main drive. An adaptive control system can be developed to solve the problem. A brief analysis of the linear controller tuners is presented in this research and a neural tuner is proposed to solve the problem. It combines an artificial neural network and a rule base. It does not require identification of the plant model or use of the test signals. In this paper the authors consider a control system for the main DC electric drive of a two-roll reversing rolling mill. A structure for the neural network is selected and the rule base is described. It defines the moments, when a controller should be adjusted (training of the neural network) as well as an appropriate learning rate. Experiments for application of the neural tuner were conducted using the rolling mill DC electric drive model, as well as a physical model of this drive. An analysis of the experimental results shows, that the neural tuner adjusts the speed of the controller effectively. The purpose of the further research is to apply the neural tuner to compensate for the disturbances acting on the rolling mill and caused by the steel cast engagement.
Keywords: controller parameters' tuning, speed P-controller, rolling mill, DC drive, adaptive control, multiloop control, neural network, rule base, neural tuner, speed loop

Acknowledgements: The work was supported by the Russian Foundation for Basic Research (grant 15-07-06092).