Training of a Classifier for Structural Component Failure based on Hybrid Simulation and Kriging

Abstract

Hybrid simulation is a tool for discovering the inner workings of a tested substructure beyond the linear regime. Hybrid simulation is conducted to reproduce the response of a prototype in scaled or real time using a hybrid model that combines physical and numerical substructures interacting with each other in a feedback loop. As a result, the tested substructure interacts with a realistic assembly subjected to a credible loading scenario. The obtained low-quantity-high-value experimental data is used to conceive and calibrate computational models for nonlinear structural analysis in the current practice. Instead, this paper extends the scope of hybrid simulation to constructing a safe/failure state classifier for the tested substructure by adaptively designing a sequence of parametrized hybrid simulations. Such a classifier is intended to compute the state of any physical-substructure-like component within system-level numerical simulations. The proposed procedure is experimentally validated for a three-degrees-of-freedom hybrid model subjected to Euler buckling.