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

Table 2

Parameter sensitivities qualitatively determined by multiple forward FEM simulations (1st row) and inverse ANN training results (quantified by 𝑅² from (9), 2nd row) with individual strain data normalization and single parameter training as shown in Figure 6(d) for 𝜎^W₀. Good training results have been obtained for the sensitive parameters, whereas nonsensitive parameters could not be satisfactorily trained. Both qualitative and quantitative sensitivity analyses have been made easy with a real-time monitoring tool of the software.

	𝜎^W0	𝜎^W1	𝜃^W0	𝜃^W1	𝜎^BMG0	𝑛^BMG	𝑇
Sensitivity	Highest	Highest	High	Lowest	Low	Lowest	Highest
𝑅^2	0.999	0.999	0.967	0.164	0.405	0.050	0.998