Development of a Neural Network Simulator for Studying the Constitutive Behavior of Structural Composite Materials

Figure 6

ANN inverse training results for the highly sensitive 𝜎^W₀ material parameter (yield strength) from the 500 FEM testing data. Figures 6(a) and 6(b) show the 𝜎^W₀ training case in which the other 6 material parameters have also been used. Figures 6(c) and 6(d) illustrate the case in which 𝜎^W₀ was trained alone. However, the whole normalization of 70 strains (1 group) over 4,500 datasets generates Figures 6(a) and 6(c), whereas Figures 6(b) and 6(d) come from the individual normalization of the 70 strain input data (70 groups).

(a) 70-80-80-7 Whole normalization for strain input layer

(b) 70-80-80-7 Individual normalization for strain input layer

(c) 70-80-80-1 Whole normalization for strain input layer

(d) 70-80-80-1 Individual normalization for strain input layer