12段クラスター型圧延機におけるディープラーニングを活用した平坦度制御技術の開発

抄録

A flatness control in a cold rolling has a significant importance for product quality and productivity. An automatic feedback flatness control based on influence coefficient is therefore commonly used in 4Hi or 6Hi rolling mills. In this kind of conventional feedback control, to determine the influence coefficients, flatness actuators are independently operated in the actual mills. It is known that effects of flatness actuators on the strip flatness changes based on the thermal crown, wear of work rolls and so on. Building and managing a model which considers such effects of all parameters are unrealistic. On the other hand, due to the difficulty of determining the influence coefficients, such flatness control is not common in 12Hi or 20Hi multi-high rolling mills. In this study, a new flatness control based on deep learning, which doesn’t need influence coefficients was proposed. In the actual mill trials, same strip flatness was obtained with fewer manually operated flatness actuators. In the considerations, it was found that the correct influence coefficients can be learned in the training phase.