Entry temperature prediction of a hot strip mill by a hybrid learning type-2 FLS

Abstract

In Hot Strip Mills, on-line estimation of rolling variables is of crucial importance in order to Set-Up the Finishing Mill, i.e. setting initial working references for the in-bar regulators, and hence fulfilling quality requirements. This paper presents the experimental results of the application of type-2 fuzzy logic systems for scale breaker entry temperature prediction in a real hot strip mill. Since in the literature only back-propagation has been proposed for type-2 fuzzy logic systems, a hybrid learning algorithm has been developed. Such algorithm is also presented. The algorithm uses back-propagation with recursive least-squares and back propagation with square-root filter methods. The systems were tested for three types of inputs: a) interval singleton b) interval type-1 non-singleton, c) interval type-2 non-singleton. The experiments were carried out for three different types of coils. Experimental results show the feasibility of the systems developed here for scale breaker entry temperature prediction. Comparison with type-1 fuzzy logic systems shows the hybrid learning type-2 fuzzy logic systems improve performance in scale breaker entry temperature prediction under the tested condition.