Abstract
This paper presents a system based on data mining and statistical modelling tools that permits the prediction of the development of oxide scale defects in high quality flat products after the steel industry’s hot strip mill process (HSM), but before the coil becomes processed on the pickling line (PL). The economic impact of the improvement provided by such a system can be valued at several million US dollars per year, because it makes it possible to downgrade materials at an early stage, avoiding additional processes like coating, etc. It also enables the speed of the PL, which is usually seen as a bottleneck in these facilities, to be increased. The learning process of the model presented here is based on automatic surface-inspection systems, as well as processing parameters at the HSM and PL to capture the essentials of the cleaning process itself, and also the main factors in scale production. The system proposed currently which is configured as a multi-agent system, is the first for this particular purpose, although the steel industry uses many other models and systems to predict other properties (e.g., mechanical properties) or the best operating parameters (e.g., forces, temperatures) for processes.
Acknowledgements
This work was supported in part by the RFCS research project ‘IRSIS’ (code RFSR-CT-2006-00036), as well as by the Spanish national research grant ‘Salvador de Madariaga’ DPI2010-0014. For these reasons, the authors express their personal appreciation to the European Commission for its support and, in particular, to the RFCS program as well as to the Spanish ‘Ministerio de Educación’.