Predictive Modelling of the Hot Rolling Process using an Integrated based Network
The rolling process is considered to be one of the most crucial operational units in the manufacture of finished and semi-finished bulk products. Predicting the final product properties is important because it can aid technologists in determining the adequate process variables required to produce a particular product with predefined characteristics. However, modelling such a process is not a trivial task, in particular when it is significantly affected by many variables and carried-out at high temperatures, the so-called hot-rolling mill process. In this research work, an integrated network based on computational intelligence is presented to model the hot-rolling mill process. In such a network, the output can be predicted via two phases. The first phase consists of a number of models that are trained using the process inputs and the target output (i.e. product characteristics). The second phase relates to utilizing the predicted outputs from the first phase as well as the target output to train another linear or nonlinear model in order to produce the final predicted output. Validated on real industrial data, it has been shown that such a modelling structure has the ability to deal with sparse data and to accurately predict the characteristics of the hot rolled products.
Index Terms - Computational Intelligence, Hot-Rolling Mill, Integrated Networks, Radial Basis Functions, Steel Sections.