Classification of Thermal Profiles in Blast Furnace Walls by Neural Networks
The wall or cooling water temperatures in the iron-making blast furnace are important sources of information about the internal conditions of the process, which cannot be measured directly because of high temperatures, mechanic wear and hostile environment. Two alternative methods of classification of wall thermal load profiles are studied: a feedforward neural network trained on manually classified temperature profiles, and an unsupervised classification using a Kohonen network. Both approaches were found to yield interesting results. In the former approach, the results can be easily explained since the characteristics of the profile classes are known, while in the latter approach the generalization properties and the ease of retraining the networks were considered advantageous. The classifiers are being implemented in the automation system of a Finnish blast furnace.
KeywordsBlast Furnace Feedforward Neural Network Unsupervised Classification Cooling Water Temperature Fuel Rate
Unable to display preview. Download preview PDF.
- A.B. Bulsari and H. Saxén. Classification of blast furnace probe temperatures using neural networks. Steel Research, 66:231–236, 1995.Google Scholar
- T. Hirata et al. Blast furnace Operationsystem using neural networks and knowledge-base. In Proceedings of the 6th International Iron and Steel Congress, pages 23–27. Nagoya, Japan, ISIJ, 1990.Google Scholar
- S. Hirose et al. Application of AI systems to Mizushima no.3 BF. In Proceedings of the 51st Ironmaking Conference, pages 163–170. Toronto, Canada, 1993.Google Scholar
- Y. Otsuka et al. Application of neural network systems to pattern recognition of blast furnace operation data. Kobelco Technology Review, 15:12–16, 1992.Google Scholar