This paper is about a project performed with Hot Rolling Dep. Of POSCO (Pohang Iron & Steel Co.) to enhance the quality of hot coil products. The quality of hot coil strip depends highly on the accuracy of coiling temperature control. Coiling temperature is the temperature of a hot iron coil strip just before it is rolled. Water is sprayed and cools down the hot coil and the control system manipulates the amount of sprayed water to control coiling temperature. To calculate the amount of water needed to get a specific coiling temperature, control system uses a mathematical model of the cooling process. This mathematical model has not been changed since it was made and as the system gets older it does not act as an accurate model of the system. A new model using recurrent neural networks and EP(Evolutionary Programming) learning algorithm is suggested in this paper to substitute the original mathematical model. Real data from past work experiences are used to train this recurrent neural networks and other real test data are used to check the validity of the new model.
