Application of a neural network concept to analyzing the expansion behavior of a monolithic refractory lining Part 1: Analysis of monolithic lining strength by nonlinear finite element method

摘要

In design analysis of refractory materials, it isnecessary for the analyst to establish an appropriate methodologyby knowing thoroughly the behavior related to inhomogeneity andpertinent nonlinearity. An alumina-magnesia compound, inwhich a reaction may occur during the service period, wouldexhibit a complicated expansion behavior such as expansion byreaction and a residual expansion, dependent on time, temperature,and heating rate.In this report, a neural network approach is introduced thatsuccessfully copes with the complicated total expansion behavior.The network was educated by being fed with a series ofexperimental isothermal expansion data, so that it can model thenonlinear expansion behavior of composite materials complicatedby chemical reactions, sintering, and thermal expansion under repeated heating.The details of the procedure are given and some simulatedexpansion results are shown in comparison with experimentalresults. The network is embedded in a finite element method andapplied to the lining of a steel ladle subjected to cyclic thermal shock.