Application of a self-learning function to an expert system for blast furnace heat control.

Abstract

A seif-learning function, in which the statistical methodsand production rules were eftectively combined, was applied to an expert system for blast furnace heat control to improve controllability of temperature and chemical composition of hot metal and to improve mentenance of the system. This self-learning function consists of a short term self-learning function and a long term self-learning function. The former has been in operation since the expert system for blast furnace heat control, which was namedBAISYS, was started and the latter has been utilized since March, 1988. In the short term self-learning function, the reference values of sensor data, hot metal temperature difterences among tapholes and rising patterns of hot metal temperature in the casting are periodically judged and automatically processed.In the long term self-learning function, the guidance to modify the three-dimensional membershipfunctions and the weight coefticients of all sensors is performed. Through the test application of the self-iearning function, the followings were confirmed.(1) Guidance by the self-leaming function is very efiective for the evaluation of weight coefficients of various sensors.(2) The three-dimensionai membershipfunctions for all sensor data, which are madeby self-learning function, are applicable.(3) The standard deviations of hot metal temperature and silicon content in hot metal have been decreased and the application ratio of the expert system has been kept at high level by the introduction of the self-learning function.

Development of a Visual Information Technique of Nonstationary Fluctuations in a Blast Furnace Process

“…[12][13][14][15][16][17] Despite the progress of such physical models and AI, and the development in the varieties of sensors and probes, however, the grasp and prediction of nonstationary phenomena in the actual blast furnaces are largely due to the experience and skills of on-site operators.…”

“…Quantitative analysis and control techniques were also developed by applying AI. [12][13][14][15][16][17] Despite the progress of such physical models and AI, and the development in the varieties of sensors and probes, however, the grasp and prediction of nonstationary phenomena in the actual blast furnaces are largely due to the experience and skills of on-site operators.One of the reasons why the identification and prediction control of nonstationary phenomena have not necessarily been automated is because the hardware was not sufficiently advanced. Since the nonstationary phenomena take place in certain radial, vertical, and peripheral zones within blast furnaces, the sampling and analysis of two-or three-dimensional data are required.…”