Community Research and Development Information Service - CORDIS

FP5

Neural networks for process condition prediction in basic oxygen steelmaking

Funded under: FP5-GROWTH

Abstract

The aim of the project was to investigate the potential application of neural networks for improvement of basic oxygen steel making (BOS) process control. Although both British Steel and Hoogovens use existing conventional process control models developed many years ago, it was recognised that they suffer from a degree of imprecision due to the effects of several process variables which are difficult to quantify. Recent advances in soft computing techniques, in particular artificial neural network (ANN) models, offered a possible solution to enhancing the accuracy, or even full replacement, of conventional models. RWTH has developed considerable expertise in the application of such techniques, and its role in the project was to develop neural network models in collaboration with the industrial partners. The feasibility of other techniques, including conventional statistics and other soft computing methods, such as fuzzy logic or case-based reasoning, was also considered.

A number of different approaches for model development were studied, including the application of ANN models at each stage of the BOS process cycle, either alone or linked to the output of the conventional models. Although some encouraging results were obtained, it was concluded that pure ANN models used in isolation offered little advantage as a direct replacement for the existing conventional models for charge balancing and sublance control. Their accuracy was found to be no better, no worse, than that of the conventional models, their long-term robustness was suspect, and they required a significant effort to develop and maintain. .

Additional information

Authors: WHITTAKER H, British Steel Ltd Swindon Technological Centre Moorgate Rotherham (UK);MULDER R, Hoogovens IJmuiden (NL);HARTWIG M, Hoogovens IJmuiden (NL);POSCHMANN M, RWTH Aachen Templergraben (DE)
Bibliographic Reference: EUR 19467 EN (2001), pp.185. Euro: 28.00
Availability: Available by EUR-OP sales agent http://www.eur-op.eu
ISBN: ISBN: 92-894-1126-0
Record Number: 200013717 / Last updated on: 2001-08-24
Category: PUBLICATION
Original language: en
Available languages: en