Diagnosis of the instability of the cooling behaviour of flat steel products through parametric characterisation, neural networks and statistics.

This paper presents a mathematical model developed by means of an analytical function whose shape depends on the values of a few parameters for the run-out table cooling which is used in hot strip mills. The system relies on a first-order differential equation for describing the temperature loss along the run-out table. Neural networks have been applied in order to find correlations between the model parameters and the steel and process variables. Then, traditional statistical techniques have been applied in order to evaluate the stability of the cooling behaviour. Numerical results obtained on an industrial database are presented and discussed.