Study On The Metamodel Based Adaptive Sampling Methods And Real-Time Hybrid Simulation

Performance based earthquake engineering requires that uncertainties in structural properties and ground motions should be explicitly accounted for.Traditional computational approach often requires accurate finite element modeling therefore does not apply to structures with parts difficult for numerical modeling.Real-time hybrid simulation(RTHS)leverages both numerical modeling and experimental testing to replicate structural response under earthquakes.Using the substructure test principle,RTHS divides a structural system into the numerical and physical substructures,where the former is analytically modeled through a finite element program and the latter is physically tested in laboratories.The method significantly reduces the experiment expense and provides the most advanced solution for large-scale structures with ratedependent behavior in size limited laboratories.At present,the real-time hybrid simulation method is in the rapid development stage.The traditional real-time hybrid simulation assumes that the substructure property is determined,so it cannot explain the structural uncertainty in the global response prediction.Aiming at adaptive sampling technology and proxy model in real-time hybrid simulation,this paper has carried out theoretical analysis,numerical simulation and experimental verification successively through the combination of metamodel and adaptive sampling.The main research contents are as follows:Adaptive experiment design is a necessary condition for quantifying uncertainty effectively through real-time hybrid simulation.In this paper,the current commonly used adaptive sampling methods are analyzed theoretically and compared experimentally.In the laboratory,real-time hybrid simulation tests are carried out on single-degree of freedom structures with self-resetting viscous damper(SC-VD).Based on the Kriging meta-model of initial samples,real-time hybrid simulation tests are carried out through adaptive sampling.The Kriging metamodel is updated with new sample points.Studies show that with a finite number of real-time hybrid simulation,different adaptive sampling methods can establish accurate global meta-models based on a finite number of real-time hybrid simulation tests to perform more accurate response estimation.In order to better use of Kriging and stochastic Kriging(SK)based on adaptive sampling,further expand the seismic mitigation capability of real-time hybrid simulation to provide insights,expand the selection range of adaptive sampling and the ability to apply to real-time hybrid simulation.Previous studies often focus on uncertainties in numerical substructures.This might be valid when the experimental substructures are repeatedly used for sequential tests.When replacement becomes necessary due to damage for experimental substructures such as fuse elements,uncertainty might be introduced which should be examined for its potential influence on selected adaptive experimental design.In order to consider both the inherent uncertainty in stochastic simulation and the external uncertainty about the unknown response surface,this study explores the stochastic Kriging technique to account for potential uncertainty due to different experimental substructures for uncertainty quantification through RTHS.A numerical example is first used to illustrate the influence of experimental substructure uncertainty.Computational simulation is further conducted RTHS of a simple structure with a self-centering viscous device to evaluate the potential of stochastic Kriging for uncertainty quantification and expand the capability of using real-time hybrid simulation to predict the uncertainty of earthquake disaster with response.