An approach to the analysis of thickness deviations in stainless steel coils based on self-organising map neural networks

Abstract

The aim of this work is to classify the sections of coils produced on a cool rolling mill that have an irregular thickness pattern, in order to achieve a homogeneous thickness in each coil. In order to do this investigation, we have employed a self-organising map (SOM) of neural networks, a new segmentation and clustering algorithm, filters to reduce the noise and, finally, a classification calculated from the difference between the value of each sample taken and the average of them all. We have introduced an alternative approach, with improvements in the segmentation and clustering steps, which has been successfully applied in an industrial production line. Some of our limitations and future areas for investigation are also included.

A Non-lineal Mathematical Model for Annealing Stainless Steel Coils

“…Data Mining tools and multivariate statistics are useful when there is a significant historical volume and good quality (Chapple, 2002). The thermal energy received by each of the coil while in the furnace can be calculated (Spinola, 2004). To do this, the temperatures applied to each coil in each area, top and bottom, are obtained from the data files where they have been continuously registered.…”

“…If we choose m2>0, temperatures higher than Ta overcompensate the annealing time, as the annealing process speed and the temperature are related. The parameter m2 reflects that fact (Spinola, 2004).…”