NEURAL NETWORK MODELS FOR THE PREDICTION OF ALUMINIUM HEATING AND ROLLING TIMES

This paper presents the artificial neural network (ANN) models used in (Ozen et al) to estimate the heating and rolling times in an aluminium process. (Ozen et al) identifies scheduling of the soaking pits/rolling mill process at an aluminium plant as an NP-Hard two-stage flow-shop problem and presents an algorithm (SPO) to synthesise a schedule that aims to increase the throughput and decrease the energy consumption. The schedule focuses on the process where groups of aluminium slabs are loaded into a number of soaking pits to be heated and after the content of a pit is soaked at the required temperature, the slabs are drawn out one by one and are rolled at a 4-high stand hot reversing mill. In order to produce a successful schedule, estimating the processing times of these two operations of heating and rolling is important. SPO uses neural network models trained by previous plant data to predict the duration of these operations. The aim of this work is to obtain better accuracy and resolution in calculation of these processing times, which are estimated manually at present.