Real-time Detection of Roll Eccentricity for Cold Strip Mills Using Multi-fraction Neural Network

摘要

In cold strip mills, roll eccentricity is one of the most important disturbance affecting thickness quality of rolled strips. Roll eccentricity is caused by both work rolls and backup rolls and is periodic and sinusoidal in nature. However, in practice the amplitude, frequency and phase of eccentricity signal are time-varying and the accuracy of detecting eccentricity is depending on measured roll force which is corrupted by hysteresis and nonlinearity effect. Motivated by time-varying and nonlinear property of detection, a novel method of detection based on neural network is proposed. In order to accomplish the real-time constraints of time variant eccentricity, this paper present a multi-fraction neural network which consists of several small-scale neural networks with identical structure connected in parallel. Eccentricity detection simulations are conducted at various conditions and the results are presented and analyzed.