Human model for gain tuning of looper control in hot strip rolling

In recent years, many control systems have been developed and applied to hot rolling in order to improve the quality of the finished products. Due to the plant characteristics and the control system performance variations over time, human experts intervene and modulate control gains to maintain and to improve the control system performance. Through a breakthrough, we can expect to attain the fully automated modulation of control gains without human intervention. A neural network model representing gain modulating actions by human was developed for a looper height controller in hot strip mills. The developed neural network model is a recurrent type neural network (RNN) which calculates the appropriate PID gains of the looper height controller based on the modification data of human operations as training data. Further, a learning algorithm for the RNN model was developed to accelerate convergence of the gain modification process and to stabilize the looper movement. The neural gain tuning model was applied to the inter-stands looper height controller in hot strip mills. The usefulness of the developed model was checked through numerical experiments. From the experimental results, it was verified that the tuning actions by humans can be realized by the model. Through its learning mechanism, the model could also cope with disturbances such as changes in roll gap. This may lead to the stabilization of threading operations of hot strip mills.