Development and implementation of a generic blast-furnace expert system

The performance of a blast-furnace can be judged by various criteria, such as reductant rate (i.e. the weight of coke and tuyere injectant required to produce 1 t of iron), consistency of product quality, measured in terms te mperature and silicon content; plant availability, and campaign life (the q uantity of iron produced before the furnace requires relining or rebuilding ). Increasing numbers of sensors mean that a large volume of data is now be ing made available to the shift controller, and this, combined with reducti ons in shift manning and the development of new data assessment techniques, makes the process an ideal candidate for an operator guidance system. Corus, the company formed by the recent merger of British Steel and Hoogove ns, operates nine blast-furnaces on four integrated sites in the United Kin gdom-two in northeast England (Redcar and Scunthorpe) and two in South Wale s (Port Talbot and Llanwern); the two furnaces at Llanwern are, however, du e to be decommissioned in July, 2001. The first attempt to install a blast-furnace expert system within British S teel was aborted in 1991, but a number of software shells were subsequently assessed by the Computing Applications Section at British Steel's Teesside Technology Centre, and G2, produced by Gensym, was selected as the most ap propriate for further developments. A system comprising data acquisition so ftware, a comprehensive relational database, an expert system developed wit h the use of G2 and a G2 database bridge that allows data to be passed betw een the database and the expert system was launched at Llanwern in October, 1996, and has operated reliably since then. A similar system was installed at Redcar, the company's largest blast-furnace, with the aim of extending campaign life by maximizing operational stability. The opportunity was take n to redesign the knowledge base, including data access, and the result was a generic shell that can be readily applied to any blast-furnace. Some of these developments were subsequently reapplied to Llanwern. More recently, a system has been developed for the Port Talbot No. 4 blast-furnace and ano ther covering the two larger furnaces at Scunthorpe is being commissioned. An adaptive principal-component analysis algorithm that accounts for normal process variations was also developed as a project within the EU-funded Fo urth Framework.