Estimation of Surface Tension of Molten Silicates Using Neural Network Computation

Abstract

Neural network computation was applied to the estimation of surface tension in ternary silicate melts. In addition, the criterion for designing the units in the middle layer of the layer-type neural network computation was discussed. It was found that the C_p-criterion modified by considering the degrees of freedom in the neural network computation was useful for determining the number of units in the middle layer, which gives an optimal estimation. The surface tension calculated by neural network computation using units determined by the C_p-criterion virtually reproduced the experimental data in molten ternary silicates with high precision.