Artificial intelligence techniques in the hot rolling of steel

In an attempt to get around the real-time impasse associated with a conventional numerical approach to predictive modelling, an integrated AI technique has been proposed and its validity has been demonstrated. Hybrid in nature, the authors´ approach combines a “bottom-up” connectionist paradigm with a top-down real-time knowledge-based system. The immediate goal was to demonstrate the application of these techniques to specific aspects of actual, albeit small scale, hot steel rolling facilities. The neural networks are trained on a mixture of experimentally gathered data and data generated from mathematical models. This project has broken new and important ground in the technology of steel processing. Neural networks predict the temperature behaviour of a hot steel slab during run-out cooling. Based on industry data, the system discussed in this paper is able to predict the final thickness, roll separation force, and the springback of the steel slab. Furthermore, taking the mill´s loading capacity into account, a hybrid real-time knowledge-base/neural network system generates the rolling schedule needed to produce a strip of steel of a specific gauge from a slab of a given composition, initial thickness and temperature