In English

Identifying, visualizing and quantifying process disturbances at SSAB Oxelösund using multivariate modelling

Henrik Rådberg
Göteborg : Chalmers tekniska högskola, 2007. 57 s.
[Examensarbete på avancerad nivå]

Modern process lines give rise to huge amounts of data which are stored in databases. Multivariate analysis comprises useful tools to grasp useful information from the datasets. In the present study principal component analysis (PCA), projection to latent structures (PLS) and hierarchical PCA has been used to create models of five process steps at a Swedish steelworks. The focus has been to identify and explain relations to quality problems in each step, both within the step itself, but also from upstream processes using hierarchical PCA. The five process steps that have been modelled are the blast furnace, desulphurization in the torpedo car, basic oxygen steelmaking in the LD–LBE-converter, secondary steelmaking in ladle and ladle furnace and, finally, continuous casting of slabs. Among the results achieved it is found that: PLS prediction of crude iron analysis from blast furnace discharge has been made with a fraction of explained variance for external validation (Q2PS) above 20% for P, Cr, Cu, Ti, CaO, SiO2, MgO and basicity. The data resolution was relatively low. Hierarchical modelling revealed correlations between the process steps, e.g. that LD-converter treatments registered as severe slopping heats have a titanium content in the incoming crude iron that is higher than average. Heats with too high phosphorous content after LD-treatment can be identified as having low silicon content in the crude iron, which makes it impossible to create the necessary slag amount for desired phosphorous cleaning effect. High sulphur content in the torpedo car demands a long treatment time. If the silicon content is low in such a batch, there is an evident risk that it will not have high enough temperature in the secondary steelmaking. Capturing reasons for quality problems during casting is difficult due to the low variation in data. The main variations exist between the steel qualities. However, the importance of casting properties such as oscillations for visual quality of the slabs, and temperature and steel analysis for slab inner quality have been recognized.

Nyckelord: Multivariate analysis, PCA, PLS, hierarchical modelling, process modelling, blast furnace, steelworks

Publikationen registrerades 2007-06-20. Den ändrades senast 2015-01-14

CPL ID: 43071

Detta är en tjänst från Chalmers bibliotek