Compensation For Time Delay In Closed-Loop Shape Control System Of Cold Rolling Mill

Flatness is one of the most important factors to evaluating the quality of a strip. The flatness control in a cold rolling mill is an important subject because of the need for improvement in cold-rolled strip quality. It, however, is a difficult problem for a conventional approach to achieve since the cold rolling process is a highly nonlinear system in which many uncertain parameters are involved and the connection among them can not be described precisely with several formulas. The automatic closed-loop flatness control is one way to solve this problem. In this control system, shape meter is always located at a distance away from the exit of the cold rolling mill in order to protect it from destroy. And this causes time delay of feedback signal, which adds great difficulties to whole system. There is not any ready theory for designing controller in time delay control system and how to do with such system is still a problem facing scientists in the field of control. So doing research on it not only has great means in theory, but also in practical application.The configuration of closed-loop shape control in 300mm 4-high reversing cold rolling mill is given firstly. As the characters of shape control instrument do great effect on the final result, electro-hydraulic bending roll system and leaning roll system are investigated pertinently. In order to get their property, a mathematic model is set up through logical analysis, which is considered as the transcendent knowledge in next identification. After-line identification method and Matlab identification tool box are used to acquire more precise model according to lots of step response data. Thus compensation algorithm can be given based on it.Traditional Smith predictor is a famous compensator for time delay system, it employs a model of the plant characterized by linear transfer function and a time delay to transform a time delay system to a delay free problem, which makes it too sensitive to model and limits its adaptability at the same time. Adaptive and self-leaning capability of Neural Network gives a new way to solve the problem. However, complex configuration and difficulty to give initial values are two stumbling blocks that affect the efficiency of Neural Network. Because of this, new Neural Network is proposed in this paper which combines the merits of traditional Smith controller and Neural Network. One is Smith predictor based on single neural network, and another is inner model control based on PIDNN(PID Neural Network) which can learn on-line. Simulation and experiment indict that these controllers can get excellent results than traditional Smith predictor and have adaptability and some intelligent.