Premium
Optimization of a BOF's lining resistance by application of artificial neural network
Author(s) -
Cai Yudong
Publication year - 1994
Publication title -
steel research
Language(s) - English
Resource type - Journals
eISSN - 1869-344X
pISSN - 0177-4832
DOI - 10.1002/srin.199401198
Subject(s) - artificial neural network , basic oxygen steelmaking , engineering , process engineering , mathematics , biological system , mechanical engineering , materials science , steelmaking , artificial intelligence , metallurgy , computer science , biology
The artificial neural network approach for optimization of lining resistance of basic oxygen furnace is presented in this paper. A group of samples has been collected to study. 33 samples are divided into two classes, i.e. good samples (class 1) providing a furnace life of more than 1000 heats, and bad samples (class 2) meaning a furnace life of less than 1000 heats. Analysis of 18 factors influencing furnace life shows that the major factors are supplementary filling amount, blowing time, melting time, and production rate. In this research, 25 samples are used as learning material, while eight samples are used as testing material for the natural network. The factors of major influence are taken as input variables. As a result, the testing rate reaches 100 %, which indicates that the model trained is reliable. In order to increase the furnace life according to the model, the partial derivative value of the output to each feature variable at the average point of class‐2 samples was calculated. The results show, that, whereas blowing and melting time had to be decreased, both the supplementary filling amount and production rate had to be increased. This result is quite consistent with the practical experience obtained in the plant. In addition, the neural network approach also has a fault‐tolerant ability, prediction and optimization speed. To sum up, the neural network approach might be referred to as an effective assistant technique for optimization of iron and steel industry.