Characterizing rate-dependent material behaviors in self-learning simulation

Abstract

Structural testing, where inhomogeneous stress distribution is introduced within the test specimen, contains far richer information on the material behavior than the conventional material testing with uniform state of stress. We present a methodology that extracts rate-dependent material behavior using load–displacement measurements from the structural test. Self-learning capability of the rate-dependent neural network material model previously proposed by the authors is used in conjunction with the methodology. Unlike other parameter optimization methods, no prior knowledge of the material is required. The model is also capable of improving its performance as further test data become available. As an illustrative example, the method is applied to capture non-linear creep behavior of superalloy.